Upload 21 files

FaceCropper.py

@@ -0,0 +1,119 @@
+import os
+from pathlib import Path
+
+from PIL import Image
+import torch
+import torch.backends.cudnn as cudnn
+from numpy import random
+
+from models.experimental import attempt_load
+from utils.datasets import LoadStreams, LoadImages
+from utils.general import (
+    check_img_size, non_max_suppression, apply_classifier, scale_coords, xyxy2xywh, plot_one_box, strip_optimizer)
+from utils.torch_utils import select_device, load_classifier, time_synchronized
+
+import gradio as gr
+import huggingface_hub
+
+from crop import crop
+
+class FaceCrop:
+    def __init__(self):
+        self.device = select_device()
+        self.half = self.device.type != 'cpu'
+        self.results = {}
+    
+    def load_dataset(self, source):
+        self.source = source
+        self.dataset = LoadImages(source)
+        print(f'Successfully load {source}')
+    
+    def load_model(self, model):
+        self.model = attempt_load(model, map_location=self.device)
+        if self.half:
+            self.model.half()
+        print(f'Successfully load model weights from {model}')
+    
+    def set_crop_config(self, target_size, mode=0, face_ratio=3, threshold=1.5):
+        self.target_size = target_size
+        self.mode = mode
+        self.face_ratio = face_ratio
+        self.threshold = threshold
+    
+    def info(self):
+        attributes = dir(self)
+        for attribute in attributes:
+            if not attribute.startswith('__') and not callable(getattr(self, attribute)):
+                value = getattr(self, attribute)
+                print(attribute, " = ", value)
+    
+    def process(self):
+        for path, img, im0s, vid_cap in self.dataset:
+            img = torch.from_numpy(img).to(self.device)
+            img = img.half() if self.half else img.float()  # uint8 to fp16/32
+            img /= 255.0  # 0 - 255 to 0.0 - 1.0
+            if img.ndimension() == 3:
+                img = img.unsqueeze(0)
+
+            # Inference
+            pred = self.model(img, augment=False)[0]
+
+            # Apply NMS
+            pred = non_max_suppression(pred)
+
+            # Process detections
+            for i, det in enumerate(pred):  # detections per image
+
+                p, s, im0 = path, '', im0s
+
+                in_path = str(Path(self.source) / Path(p).name)
+                
+                #txt_path = str(Path(out) / Path(p).stem)
+                s += '%gx%g ' % img.shape[2:]  # print string
+                gn = torch.tensor(im0.shape)[[1, 0, 1, 0]]  # normalization gain whwh
+
+                if det is not None and len(det):
+                    # Rescale boxes from img_size to im0 size
+                    det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
+
+                    # Write results
+                    ind = 0
+                    for *xyxy, conf, cls in det:
+                        if conf > 0.6:  # Write to file
+                            out_path = os.path.join(str(Path(self.out_folder)), Path(p).name.replace('.', '_'+str(ind)+'.'))
+                            
+                            x, y, w, h = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist()
+                            self.results[ind] = crop(in_path, (x, y), out_path, mode=self.mode, size=self.target_size, box=(w, h), face_ratio=self.face_ratio, shreshold=self.threshold)
+
+                            ind += 1
+                            
+def run(img, mode, width, height):
+    face_crop_pipeline.set_crop_config(mode=mode, target_size=(width,height))
+    face_crop_pipeline.process
+    return face_crop_pipeline.results[0]
+
+if __name__ == '__main__':
+    model_path = huggingface_hub.hf_hub_download("Carzit/yolo5x_anime", "yolo5x_anime.pt")
+    face_crop_pipeline = FaceCrop()
+    face_crop_pipeline.load_model(model_path)
+    
+
+    app = gr.Blocks()
+    with app:
+        gr.Markdown("# Anime Face Crop\n\n"
+                    "![visitor badge](https://visitor-badge.glitch.me/badge?page_id=skytnt.animeseg)\n\n"
+                    "demo for [https://github.com/SkyTNT/anime-segmentation/](https://github.com/SkyTNT/anime-segmentation/)")
+        with gr.Row():
+            input_img = gr.Image(label="input image")
+            output_img = gr.Image(label="result", image_mode="RGB")
+            crop_mode = gr.Dropdown([0, 1, 2, 3], label="Crop Mode", info="0:Auto; 1:No Scale; 2:Full Screen; 3:Fixed Face Ratio")
+            tgt_width = gr.Slider(10, 2048, value=512, label="Width")
+            tgt_height = gr.Slider(10, 2048, value=512, label="Height")
+
+            run_btn = gr.Button(variant="primary")
+
+        run_btn.click(run, [input_img, crop_mode, tgt_width, tgt_height], [output_img])
+    app.launch()
+
+
+    

crop.py

@@ -0,0 +1,84 @@
+import os
+from pathlib import Path
+from PIL import Image
+import numpy as np
+
+def crop(img, point, mode=0, size=(512, 512), box=None, face_ratio=3, shreshold=1.5):
+    img_width, img_height = img.size
+    tgt_width, tgt_height = size
+    point = (point[0]*img_width, point[1]*img_height)
+
+    # mode 0 : automatic
+    if mode == 0:
+        if box is None:
+            raise RuntimeError('face bax parameter expected: missing box=(width, height)')
+        if img_width < tgt_width or img_height < tgt_height:
+            mode = 1
+        elif face_ratio ** 2 * shreshold ** 2 * box[0] * box[1] * img_width * img_height < tgt_width * tgt_height:
+            mode = 2
+        else:
+            mode = 3
+
+    # mode 1 : no scale
+    if mode == 1:
+        pass
+
+    # mode 2 : full screen - crop as largr as possible
+    if mode == 2:
+        if tgt_width/img_width > tgt_height/img_height:
+            r = tgt_height / tgt_width
+            tgt_width = img_width
+            tgt_height = round(tgt_width * r)
+        else:
+            r = tgt_width / tgt_height
+            tgt_height = img_height
+            tgt_width = round(tgt_height * r)  
+    
+    # mode 3 : fixed face ratio
+    if mode == 3:
+        if box is None:
+            raise RuntimeError('face bax parameter expected: missing box=(width, height)')
+        box_width = box[0] * img_height
+        box_height = box[1] * img_height
+        if box_width/tgt_width > box_height/tgt_width:
+            r = tgt_height / tgt_width
+            tgt_width = round(box_width * face_ratio)
+            tgt_height = round(tgt_width * r)
+        else:
+            r = tgt_width / tgt_height
+            tgt_height = round(box_height * face_ratio)
+            tgt_width = round(tgt_height * r) 
+
+
+    # upscale raw image if target size is over raw image size
+    if img_width < tgt_width or img_height < tgt_height:
+        if img_width < img_height:
+            img_height = round(tgt_width * img_height / img_width)
+            img_width = tgt_width
+            img = img.resize((img_width, img_height))
+        else:
+            img_width = round(tgt_height * img_width / img_height)
+            img_height = tgt_height
+            img = img.resize((img_width, img_height)) 
+
+    left = point[0] - tgt_width // 2
+    top = point[1] - tgt_height // 2
+    right = point[0] + tgt_width // 2
+    bottom = point[1] + tgt_height // 2
+
+    if left < 0:
+        right -= left
+        left = 0
+    if right > img_width:
+        left -= (right-img_width)
+        right = img_width
+    if top < 0:
+        bottom -= top
+        top = 0
+    if bottom > img_height:
+        top -= (bottom-img_height)
+        bottom = img_height
+
+    cropped_img = img.crop((left, top, right, bottom))
+    cropped_img = cropped_img.resize(size)
+    return np.ndarray(cropped_img)

models/common.py

@@ -0,0 +1,118 @@
+# This file contains modules common to various models
+import math
+
+import torch
+import torch.nn as nn
+
+
+def autopad(k, p=None):  # kernel, padding
+    # Pad to 'same'
+    if p is None:
+        p = k // 2 if isinstance(k, int) else [x // 2 for x in k]  # auto-pad
+    return p
+
+
+def DWConv(c1, c2, k=1, s=1, act=True):
+    # Depthwise convolution
+    return Conv(c1, c2, k, s, g=math.gcd(c1, c2), act=act)
+
+
+class Conv(nn.Module):
+    # Standard convolution
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
+        super(Conv, self).__init__()
+        self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)
+        self.bn = nn.BatchNorm2d(c2)
+        self.act = nn.LeakyReLU(0.1, inplace=True) if act else nn.Identity()
+
+    def forward(self, x):
+        return self.act(self.bn(self.conv(x)))
+
+    def fuseforward(self, x):
+        return self.act(self.conv(x))
+
+
+class Bottleneck(nn.Module):
+    # Standard bottleneck
+    def __init__(self, c1, c2, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, shortcut, groups, expansion
+        super(Bottleneck, self).__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c_, c2, 3, 1, g=g)
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class BottleneckCSP(nn.Module):
+    # CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+        super(BottleneckCSP, self).__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
+        self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
+        self.cv4 = Conv(2 * c_, c2, 1, 1)
+        self.bn = nn.BatchNorm2d(2 * c_)  # applied to cat(cv2, cv3)
+        self.act = nn.LeakyReLU(0.1, inplace=True)
+        self.m = nn.Sequential(*[Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)])
+
+    def forward(self, x):
+        y1 = self.cv3(self.m(self.cv1(x)))
+        y2 = self.cv2(x)
+        return self.cv4(self.act(self.bn(torch.cat((y1, y2), dim=1))))
+
+
+class SPP(nn.Module):
+    # Spatial pyramid pooling layer used in YOLOv3-SPP
+    def __init__(self, c1, c2, k=(5, 9, 13)):
+        super(SPP, self).__init__()
+        c_ = c1 // 2  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1)
+        self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k])
+
+    def forward(self, x):
+        x = self.cv1(x)
+        return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1))
+
+
+class Focus(nn.Module):
+    # Focus wh information into c-space
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True):  # ch_in, ch_out, kernel, stride, padding, groups
+        super(Focus, self).__init__()
+        self.conv = Conv(c1 * 4, c2, k, s, p, g, act)
+
+    def forward(self, x):  # x(b,c,w,h) -> y(b,4c,w/2,h/2)
+        return self.conv(torch.cat([x[..., ::2, ::2], x[..., 1::2, ::2], x[..., ::2, 1::2], x[..., 1::2, 1::2]], 1))
+
+
+class Concat(nn.Module):
+    # Concatenate a list of tensors along dimension
+    def __init__(self, dimension=1):
+        super(Concat, self).__init__()
+        self.d = dimension
+
+    def forward(self, x):
+        return torch.cat(x, self.d)
+
+
+class Flatten(nn.Module):
+    # Use after nn.AdaptiveAvgPool2d(1) to remove last 2 dimensions
+    @staticmethod
+    def forward(x):
+        return x.view(x.size(0), -1)
+
+
+class Classify(nn.Module):
+    # Classification head, i.e. x(b,c1,20,20) to x(b,c2)
+    def __init__(self, c1, c2, k=1, s=1, p=None, g=1):  # ch_in, ch_out, kernel, stride, padding, groups
+        super(Classify, self).__init__()
+        self.aap = nn.AdaptiveAvgPool2d(1)  # to x(b,c1,1,1)
+        self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False)  # to x(b,c2,1,1)
+        self.flat = Flatten()
+
+    def forward(self, x):
+        z = torch.cat([self.aap(y) for y in (x if isinstance(x, list) else [x])], 1)  # cat if list
+        return self.flat(self.conv(z))  # flatten to x(b,c2)

models/experimental.py

@@ -0,0 +1,145 @@
+# This file contains experimental modules
+
+import numpy as np
+import torch
+import torch.nn as nn
+
+from models.common import Conv, DWConv
+from utils.google_utils import attempt_download
+
+
+class CrossConv(nn.Module):
+    # Cross Convolution Downsample
+    def __init__(self, c1, c2, k=3, s=1, g=1, e=1.0, shortcut=False):
+        # ch_in, ch_out, kernel, stride, groups, expansion, shortcut
+        super(CrossConv, self).__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, (1, k), (1, s))
+        self.cv2 = Conv(c_, c2, (k, 1), (s, 1), g=g)
+        self.add = shortcut and c1 == c2
+
+    def forward(self, x):
+        return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x))
+
+
+class C3(nn.Module):
+    # Cross Convolution CSP
+    def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansion
+        super(C3, self).__init__()
+        c_ = int(c2 * e)  # hidden channels
+        self.cv1 = Conv(c1, c_, 1, 1)
+        self.cv2 = nn.Conv2d(c1, c_, 1, 1, bias=False)
+        self.cv3 = nn.Conv2d(c_, c_, 1, 1, bias=False)
+        self.cv4 = Conv(2 * c_, c2, 1, 1)
+        self.bn = nn.BatchNorm2d(2 * c_)  # applied to cat(cv2, cv3)
+        self.act = nn.LeakyReLU(0.1, inplace=True)
+        self.m = nn.Sequential(*[CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)])
+
+    def forward(self, x):
+        y1 = self.cv3(self.m(self.cv1(x)))
+        y2 = self.cv2(x)
+        return self.cv4(self.act(self.bn(torch.cat((y1, y2), dim=1))))
+
+
+class Sum(nn.Module):
+    # Weighted sum of 2 or more layers https://arxiv.org/abs/1911.09070
+    def __init__(self, n, weight=False):  # n: number of inputs
+        super(Sum, self).__init__()
+        self.weight = weight  # apply weights boolean
+        self.iter = range(n - 1)  # iter object
+        if weight:
+            self.w = nn.Parameter(-torch.arange(1., n) / 2, requires_grad=True)  # layer weights
+
+    def forward(self, x):
+        y = x[0]  # no weight
+        if self.weight:
+            w = torch.sigmoid(self.w) * 2
+            for i in self.iter:
+                y = y + x[i + 1] * w[i]
+        else:
+            for i in self.iter:
+                y = y + x[i + 1]
+        return y
+
+
+class GhostConv(nn.Module):
+    # Ghost Convolution https://github.com/huawei-noah/ghostnet
+    def __init__(self, c1, c2, k=1, s=1, g=1, act=True):  # ch_in, ch_out, kernel, stride, groups
+        super(GhostConv, self).__init__()
+        c_ = c2 // 2  # hidden channels
+        self.cv1 = Conv(c1, c_, k, s, g, act)
+        self.cv2 = Conv(c_, c_, 5, 1, c_, act)
+
+    def forward(self, x):
+        y = self.cv1(x)
+        return torch.cat([y, self.cv2(y)], 1)
+
+
+class GhostBottleneck(nn.Module):
+    # Ghost Bottleneck https://github.com/huawei-noah/ghostnet
+    def __init__(self, c1, c2, k, s):
+        super(GhostBottleneck, self).__init__()
+        c_ = c2 // 2
+        self.conv = nn.Sequential(GhostConv(c1, c_, 1, 1),  # pw
+                                  DWConv(c_, c_, k, s, act=False) if s == 2 else nn.Identity(),  # dw
+                                  GhostConv(c_, c2, 1, 1, act=False))  # pw-linear
+        self.shortcut = nn.Sequential(DWConv(c1, c1, k, s, act=False),
+                                      Conv(c1, c2, 1, 1, act=False)) if s == 2 else nn.Identity()
+
+    def forward(self, x):
+        return self.conv(x) + self.shortcut(x)
+
+
+class MixConv2d(nn.Module):
+    # Mixed Depthwise Conv https://arxiv.org/abs/1907.09595
+    def __init__(self, c1, c2, k=(1, 3), s=1, equal_ch=True):
+        super(MixConv2d, self).__init__()
+        groups = len(k)
+        if equal_ch:  # equal c_ per group
+            i = torch.linspace(0, groups - 1E-6, c2).floor()  # c2 indices
+            c_ = [(i == g).sum() for g in range(groups)]  # intermediate channels
+        else:  # equal weight.numel() per group
+            b = [c2] + [0] * groups
+            a = np.eye(groups + 1, groups, k=-1)
+            a -= np.roll(a, 1, axis=1)
+            a *= np.array(k) ** 2
+            a[0] = 1
+            c_ = np.linalg.lstsq(a, b, rcond=None)[0].round()  # solve for equal weight indices, ax = b
+
+        self.m = nn.ModuleList([nn.Conv2d(c1, int(c_[g]), k[g], s, k[g] // 2, bias=False) for g in range(groups)])
+        self.bn = nn.BatchNorm2d(c2)
+        self.act = nn.LeakyReLU(0.1, inplace=True)
+
+    def forward(self, x):
+        return x + self.act(self.bn(torch.cat([m(x) for m in self.m], 1)))
+
+
+class Ensemble(nn.ModuleList):
+    # Ensemble of models
+    def __init__(self):
+        super(Ensemble, self).__init__()
+
+    def forward(self, x, augment=False):
+        y = []
+        for module in self:
+            y.append(module(x, augment)[0])
+        # y = torch.stack(y).max(0)[0]  # max ensemble
+        # y = torch.cat(y, 1)  # nms ensemble
+        y = torch.stack(y).mean(0)  # mean ensemble
+        return y, None  # inference, train output
+
+
+def attempt_load(weights, map_location=None):
+    # Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a
+    model = Ensemble()
+    for w in weights if isinstance(weights, list) else [weights]:
+        attempt_download(w)
+        model.append(torch.load(w, map_location=map_location)['model'].float().fuse().eval())  # load FP32 model
+
+    if len(model) == 1:
+        return model[-1]  # return model
+    else:
+        print('Ensemble created with %s\n' % weights)
+        for k in ['names', 'stride']:
+            setattr(model, k, getattr(model[-1], k))
+        return model  # return ensemble

models/export.py

@@ -0,0 +1,74 @@
+"""Exports a YOLOv5 *.pt model to ONNX and TorchScript formats
+
+Usage:
+    $ export PYTHONPATH="$PWD" && python models/export.py --weights ./weights/yolov5s.pt --img 640 --batch 1
+"""
+
+import argparse
+
+import torch
+
+from utils.google_utils import attempt_download
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--weights', type=str, default='./yolov5s.pt', help='weights path')
+    parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='image size')
+    parser.add_argument('--batch-size', type=int, default=1, help='batch size')
+    opt = parser.parse_args()
+    opt.img_size *= 2 if len(opt.img_size) == 1 else 1  # expand
+    print(opt)
+
+    # Input
+    img = torch.zeros((opt.batch_size, 3, *opt.img_size))  # image size(1,3,320,192) iDetection
+
+    # Load PyTorch model
+    attempt_download(opt.weights)
+    model = torch.load(opt.weights, map_location=torch.device('cpu'))['model'].float()
+    model.eval()
+    model.model[-1].export = True  # set Detect() layer export=True
+    y = model(img)  # dry run
+
+    # TorchScript export
+    try:
+        print('\nStarting TorchScript export with torch %s...' % torch.__version__)
+        f = opt.weights.replace('.pt', '.torchscript.pt')  # filename
+        ts = torch.jit.trace(model, img)
+        ts.save(f)
+        print('TorchScript export success, saved as %s' % f)
+    except Exception as e:
+        print('TorchScript export failure: %s' % e)
+
+    # ONNX export
+    try:
+        import onnx
+
+        print('\nStarting ONNX export with onnx %s...' % onnx.__version__)
+        f = opt.weights.replace('.pt', '.onnx')  # filename
+        model.fuse()  # only for ONNX
+        torch.onnx.export(model, img, f, verbose=False, opset_version=12, input_names=['images'],
+                          output_names=['classes', 'boxes'] if y is None else ['output'])
+
+        # Checks
+        onnx_model = onnx.load(f)  # load onnx model
+        onnx.checker.check_model(onnx_model)  # check onnx model
+        print(onnx.helper.printable_graph(onnx_model.graph))  # print a human readable model
+        print('ONNX export success, saved as %s' % f)
+    except Exception as e:
+        print('ONNX export failure: %s' % e)
+
+    # CoreML export
+    try:
+        import coremltools as ct
+
+        print('\nStarting CoreML export with coremltools %s...' % ct.__version__)
+        # convert model from torchscript and apply pixel scaling as per detect.py
+        model = ct.convert(ts, inputs=[ct.ImageType(name='images', shape=img.shape, scale=1 / 255.0, bias=[0, 0, 0])])
+        f = opt.weights.replace('.pt', '.mlmodel')  # filename
+        model.save(f)
+        print('CoreML export success, saved as %s' % f)
+    except Exception as e:
+        print('CoreML export failure: %s' % e)
+
+    # Finish
+    print('\nExport complete. Visualize with https://github.com/lutzroeder/netron.')

models/hub/yolov3-spp.yaml

@@ -0,0 +1,51 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# darknet53 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Conv, [32, 3, 1]],  # 0
+   [-1, 1, Conv, [64, 3, 2]],  # 1-P1/2
+   [-1, 1, Bottleneck, [64]],
+   [-1, 1, Conv, [128, 3, 2]],  # 3-P2/4
+   [-1, 2, Bottleneck, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 5-P3/8
+   [-1, 8, Bottleneck, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 7-P4/16
+   [-1, 8, Bottleneck, [512]],
+   [-1, 1, Conv, [1024, 3, 2]], # 9-P5/32
+   [-1, 4, Bottleneck, [1024]],  # 10
+  ]
+
+# YOLOv3-SPP head
+head:
+  [[-1, 1, Bottleneck, [1024, False]],
+   [-1, 1, SPP, [512, [5, 9, 13]]],
+   [-1, 1, Conv, [1024, 3, 1]],
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, Conv, [1024, 3, 1]],  # 15 (P5/32-large)
+
+   [-2, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 8], 1, Concat, [1]],  # cat backbone P4
+   [-1, 1, Bottleneck, [512, False]],
+   [-1, 1, Bottleneck, [512, False]],
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, Conv, [512, 3, 1]],  # 22 (P4/16-medium)
+
+   [-2, 1, Conv, [128, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P3
+   [-1, 1, Bottleneck, [256, False]],
+   [-1, 2, Bottleneck, [256, False]],  # 27 (P3/8-small)
+
+   [[27, 22, 15], 1, Detect, [nc, anchors]],   # Detect(P3, P4, P5)
+  ]

models/hub/yolov5-fpn.yaml

@@ -0,0 +1,42 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Focus, [64, 3]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, Bottleneck, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 9, BottleneckCSP, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, BottleneckCSP, [512]],
+   [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
+   [-1, 1, SPP, [1024, [5, 9, 13]]],
+   [-1, 6, BottleneckCSP, [1024]],  # 9
+  ]
+
+# YOLOv5 FPN head
+head:
+  [[-1, 3, BottleneckCSP, [1024, False]],  # 10 (P5/32-large)
+
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 1, Conv, [512, 1, 1]],
+   [-1, 3, BottleneckCSP, [512, False]],  # 14 (P4/16-medium)
+
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 3, BottleneckCSP, [256, False]],  # 18 (P3/8-small)
+
+   [[18, 14, 10], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/hub/yolov5-panet.yaml

@@ -0,0 +1,48 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [116,90, 156,198, 373,326]  # P5/32
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [10,13, 16,30, 33,23]  # P3/8
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Focus, [64, 3]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, BottleneckCSP, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 9, BottleneckCSP, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, BottleneckCSP, [512]],
+   [-1, 1, Conv, [1024, 3, 2]], # 7-P5/32
+   [-1, 1, SPP, [1024, [5, 9, 13]]],
+   [-1, 3, BottleneckCSP, [1024, False]],  # 9
+  ]
+
+# YOLOv5 PANet head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, BottleneckCSP, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, BottleneckCSP, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P5, P4, P3)
+  ]

models/yolo.py

@@ -0,0 +1,259 @@
+import argparse
+import math
+from copy import deepcopy
+from pathlib import Path
+
+import torch
+import torch.nn as nn
+
+from models.common import Conv, Bottleneck, SPP, DWConv, Focus, BottleneckCSP, Concat
+from models.experimental import MixConv2d, CrossConv, C3
+from utils.general import check_anchor_order, make_divisible, check_file
+from utils.torch_utils import (
+    time_synchronized, fuse_conv_and_bn, model_info, scale_img, initialize_weights, select_device)
+
+
+class Detect(nn.Module):
+    def __init__(self, nc=80, anchors=(), ch=()):  # detection layer
+        super(Detect, self).__init__()
+        self.stride = None  # strides computed during build
+        self.nc = nc  # number of classes
+        self.no = nc + 5  # number of outputs per anchor
+        self.nl = len(anchors)  # number of detection layers
+        self.na = len(anchors[0]) // 2  # number of anchors
+        self.grid = [torch.zeros(1)] * self.nl  # init grid
+        a = torch.tensor(anchors).float().view(self.nl, -1, 2)
+        self.register_buffer('anchors', a)  # shape(nl,na,2)
+        self.register_buffer('anchor_grid', a.clone().view(self.nl, 1, -1, 1, 1, 2))  # shape(nl,1,na,1,1,2)
+        self.m = nn.ModuleList(nn.Conv2d(x, self.no * self.na, 1) for x in ch)  # output conv
+        self.export = False  # onnx export
+
+    def forward(self, x):
+        # x = x.copy()  # for profiling
+        z = []  # inference output
+        self.training |= self.export
+        for i in range(self.nl):
+            x[i] = self.m[i](x[i])  # conv
+            bs, _, ny, nx = x[i].shape  # x(bs,255,20,20) to x(bs,3,20,20,85)
+            x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
+
+            if not self.training:  # inference
+                if self.grid[i].shape[2:4] != x[i].shape[2:4]:
+                    self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
+
+                y = x[i].sigmoid()
+                y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i].to(x[i].device)) * self.stride[i]  # xy
+                y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i]  # wh
+                z.append(y.view(bs, -1, self.no))
+
+        return x if self.training else (torch.cat(z, 1), x)
+
+    @staticmethod
+    def _make_grid(nx=20, ny=20):
+        yv, xv = torch.meshgrid([torch.arange(ny), torch.arange(nx)])
+        return torch.stack((xv, yv), 2).view((1, 1, ny, nx, 2)).float()
+
+
+class Model(nn.Module):
+    def __init__(self, cfg='yolov5s.yaml', ch=3, nc=None):  # model, input channels, number of classes
+        super(Model, self).__init__()
+        if isinstance(cfg, dict):
+            self.yaml = cfg  # model dict
+        else:  # is *.yaml
+            import yaml  # for torch hub
+            self.yaml_file = Path(cfg).name
+            with open(cfg) as f:
+                self.yaml = yaml.load(f, Loader=yaml.FullLoader)  # model dict
+
+        # Define model
+        if nc and nc != self.yaml['nc']:
+            print('Overriding %s nc=%g with nc=%g' % (cfg, self.yaml['nc'], nc))
+            self.yaml['nc'] = nc  # override yaml value
+        self.model, self.save = parse_model(deepcopy(self.yaml), ch=[ch])  # model, savelist, ch_out
+        # print([x.shape for x in self.forward(torch.zeros(1, ch, 64, 64))])
+
+        # Build strides, anchors
+        m = self.model[-1]  # Detect()
+        if isinstance(m, Detect):
+            s = 128  # 2x min stride
+            m.stride = torch.tensor([s / x.shape[-2] for x in self.forward(torch.zeros(1, ch, s, s))])  # forward
+            m.anchors /= m.stride.view(-1, 1, 1)
+            check_anchor_order(m)
+            self.stride = m.stride
+            self._initialize_biases()  # only run once
+            # print('Strides: %s' % m.stride.tolist())
+
+        # Init weights, biases
+        initialize_weights(self)
+        self.info()
+        print('')
+
+    def forward(self, x, augment=False, profile=False):
+        if augment:
+            img_size = x.shape[-2:]  # height, width
+            s = [1, 0.83, 0.67]  # scales
+            f = [None, 3, None]  # flips (2-ud, 3-lr)
+            y = []  # outputs
+            for si, fi in zip(s, f):
+                xi = scale_img(x.flip(fi) if fi else x, si)
+                yi = self.forward_once(xi)[0]  # forward
+                # cv2.imwrite('img%g.jpg' % s, 255 * xi[0].numpy().transpose((1, 2, 0))[:, :, ::-1])  # save
+                yi[..., :4] /= si  # de-scale
+                if fi == 2:
+                    yi[..., 1] = img_size[0] - yi[..., 1]  # de-flip ud
+                elif fi == 3:
+                    yi[..., 0] = img_size[1] - yi[..., 0]  # de-flip lr
+                y.append(yi)
+            return torch.cat(y, 1), None  # augmented inference, train
+        else:
+            return self.forward_once(x, profile)  # single-scale inference, train
+
+    def forward_once(self, x, profile=False):
+        y, dt = [], []  # outputs
+        for m in self.model:
+            if m.f != -1:  # if not from previous layer
+                x = y[m.f] if isinstance(m.f, int) else [x if j == -1 else y[j] for j in m.f]  # from earlier layers
+
+            if profile:
+                try:
+                    import thop
+                    o = thop.profile(m, inputs=(x,), verbose=False)[0] / 1E9 * 2  # FLOPS
+                except:
+                    o = 0
+                t = time_synchronized()
+                for _ in range(10):
+                    _ = m(x)
+                dt.append((time_synchronized() - t) * 100)
+                print('%10.1f%10.0f%10.1fms %-40s' % (o, m.np, dt[-1], m.type))
+
+            x = m(x)  # run
+            y.append(x if m.i in self.save else None)  # save output
+
+        if profile:
+            print('%.1fms total' % sum(dt))
+        return x
+
+    def _initialize_biases(self, cf=None):  # initialize biases into Detect(), cf is class frequency
+        # cf = torch.bincount(torch.tensor(np.concatenate(dataset.labels, 0)[:, 0]).long(), minlength=nc) + 1.
+        m = self.model[-1]  # Detect() module
+        for mi, s in zip(m.m, m.stride):  # from
+            b = mi.bias.view(m.na, -1)  # conv.bias(255) to (3,85)
+            b[:, 4] += math.log(8 / (640 / s) ** 2)  # obj (8 objects per 640 image)
+            b[:, 5:] += math.log(0.6 / (m.nc - 0.99)) if cf is None else torch.log(cf / cf.sum())  # cls
+            mi.bias = torch.nn.Parameter(b.view(-1), requires_grad=True)
+
+    def _print_biases(self):
+        m = self.model[-1]  # Detect() module
+        for mi in m.m:  # from
+            b = mi.bias.detach().view(m.na, -1).T  # conv.bias(255) to (3,85)
+            print(('%6g Conv2d.bias:' + '%10.3g' * 6) % (mi.weight.shape[1], *b[:5].mean(1).tolist(), b[5:].mean()))
+
+    # def _print_weights(self):
+    #     for m in self.model.modules():
+    #         if type(m) is Bottleneck:
+    #             print('%10.3g' % (m.w.detach().sigmoid() * 2))  # shortcut weights
+
+    def fuse(self):  # fuse model Conv2d() + BatchNorm2d() layers
+        print('Fusing layers... ', end='')
+        for m in self.model.modules():
+            if type(m) is Conv:
+                m._non_persistent_buffers_set = set()  # pytorch 1.6.0 compatability
+                m.conv = fuse_conv_and_bn(m.conv, m.bn)  # update conv
+                m.bn = None  # remove batchnorm
+                m.forward = m.fuseforward  # update forward
+        self.info()
+        return self
+
+    def info(self):  # print model information
+        model_info(self)
+
+
+def parse_model(d, ch):  # model_dict, input_channels(3)
+    print('\n%3s%18s%3s%10s  %-40s%-30s' % ('', 'from', 'n', 'params', 'module', 'arguments'))
+    anchors, nc, gd, gw = d['anchors'], d['nc'], d['depth_multiple'], d['width_multiple']
+    na = (len(anchors[0]) // 2) if isinstance(anchors, list) else anchors  # number of anchors
+    no = na * (nc + 5)  # number of outputs = anchors * (classes + 5)
+
+    layers, save, c2 = [], [], ch[-1]  # layers, savelist, ch out
+    for i, (f, n, m, args) in enumerate(d['backbone'] + d['head']):  # from, number, module, args
+        m = eval(m) if isinstance(m, str) else m  # eval strings
+        for j, a in enumerate(args):
+            try:
+                args[j] = eval(a) if isinstance(a, str) else a  # eval strings
+            except:
+                pass
+
+        n = max(round(n * gd), 1) if n > 1 else n  # depth gain
+        if m in [nn.Conv2d, Conv, Bottleneck, SPP, DWConv, MixConv2d, Focus, CrossConv, BottleneckCSP, C3]:
+            c1, c2 = ch[f], args[0]
+
+            # Normal
+            # if i > 0 and args[0] != no:  # channel expansion factor
+            #     ex = 1.75  # exponential (default 2.0)
+            #     e = math.log(c2 / ch[1]) / math.log(2)
+            #     c2 = int(ch[1] * ex ** e)
+            # if m != Focus:
+
+            c2 = make_divisible(c2 * gw, 8) if c2 != no else c2
+
+            # Experimental
+            # if i > 0 and args[0] != no:  # channel expansion factor
+            #     ex = 1 + gw  # exponential (default 2.0)
+            #     ch1 = 32  # ch[1]
+            #     e = math.log(c2 / ch1) / math.log(2)  # level 1-n
+            #     c2 = int(ch1 * ex ** e)
+            # if m != Focus:
+            #     c2 = make_divisible(c2, 8) if c2 != no else c2
+
+            args = [c1, c2, *args[1:]]
+            if m in [BottleneckCSP, C3]:
+                args.insert(2, n)
+                n = 1
+        elif m is nn.BatchNorm2d:
+            args = [ch[f]]
+        elif m is Concat:
+            c2 = sum([ch[-1 if x == -1 else x + 1] for x in f])
+        elif m is Detect:
+            args.append([ch[x + 1] for x in f])
+            if isinstance(args[1], int):  # number of anchors
+                args[1] = [list(range(args[1] * 2))] * len(f)
+        else:
+            c2 = ch[f]
+
+        m_ = nn.Sequential(*[m(*args) for _ in range(n)]) if n > 1 else m(*args)  # module
+        t = str(m)[8:-2].replace('__main__.', '')  # module type
+        np = sum([x.numel() for x in m_.parameters()])  # number params
+        m_.i, m_.f, m_.type, m_.np = i, f, t, np  # attach index, 'from' index, type, number params
+        print('%3s%18s%3s%10.0f  %-40s%-30s' % (i, f, n, np, t, args))  # print
+        save.extend(x % i for x in ([f] if isinstance(f, int) else f) if x != -1)  # append to savelist
+        layers.append(m_)
+        ch.append(c2)
+    return nn.Sequential(*layers), sorted(save)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--cfg', type=str, default='yolov5s.yaml', help='model.yaml')
+    parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
+    opt = parser.parse_args()
+    opt.cfg = check_file(opt.cfg)  # check file
+    device = select_device(opt.device)
+
+    # Create model
+    model = Model(opt.cfg).to(device)
+    model.train()
+
+    # Profile
+    # img = torch.rand(8 if torch.cuda.is_available() else 1, 3, 640, 640).to(device)
+    # y = model(img, profile=True)
+
+    # ONNX export
+    # model.model[-1].export = True
+    # torch.onnx.export(model, img, opt.cfg.replace('.yaml', '.onnx'), verbose=True, opset_version=11)
+
+    # Tensorboard
+    # from torch.utils.tensorboard import SummaryWriter
+    # tb_writer = SummaryWriter()
+    # print("Run 'tensorboard --logdir=models/runs' to view tensorboard at http://localhost:6006/")
+    # tb_writer.add_graph(model.model, img)  # add model to tensorboard
+    # tb_writer.add_image('test', img[0], dataformats='CWH')  # add model to tensorboard

models/yolov5l.yaml

@@ -0,0 +1,48 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 1.0  # model depth multiple
+width_multiple: 1.0  # layer channel multiple
+
+# anchors
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Focus, [64, 3]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, BottleneckCSP, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 9, BottleneckCSP, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, BottleneckCSP, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 1, SPP, [1024, [5, 9, 13]]],
+   [-1, 3, BottleneckCSP, [1024, False]],  # 9
+  ]
+
+# YOLOv5 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, BottleneckCSP, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, BottleneckCSP, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/yolov5m.yaml

@@ -0,0 +1,48 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 0.67  # model depth multiple
+width_multiple: 0.75  # layer channel multiple
+
+# anchors
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Focus, [64, 3]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, BottleneckCSP, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 9, BottleneckCSP, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, BottleneckCSP, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 1, SPP, [1024, [5, 9, 13]]],
+   [-1, 3, BottleneckCSP, [1024, False]],  # 9
+  ]
+
+# YOLOv5 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, BottleneckCSP, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, BottleneckCSP, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/yolov5s.yaml

@@ -0,0 +1,48 @@
+# parameters
+nc: 80  # number of classes
+depth_multiple: 0.33  # model depth multiple
+width_multiple: 0.50  # layer channel multiple
+
+# anchors
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Focus, [64, 3]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, BottleneckCSP, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 9, BottleneckCSP, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, BottleneckCSP, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 1, SPP, [1024, [5, 9, 13]]],
+   [-1, 3, BottleneckCSP, [1024, False]],  # 9
+  ]
+
+# YOLOv5 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, BottleneckCSP, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, BottleneckCSP, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

models/yolov5x.yaml

@@ -0,0 +1,48 @@
+# parameters
+nc: 1  # number of classes
+depth_multiple: 1.33  # model depth multiple
+width_multiple: 1.25  # layer channel multiple
+
+# anchors
+anchors:
+  - [10,13, 16,30, 33,23]  # P3/8
+  - [30,61, 62,45, 59,119]  # P4/16
+  - [116,90, 156,198, 373,326]  # P5/32
+
+# YOLOv5 backbone
+backbone:
+  # [from, number, module, args]
+  [[-1, 1, Focus, [64, 3]],  # 0-P1/2
+   [-1, 1, Conv, [128, 3, 2]],  # 1-P2/4
+   [-1, 3, BottleneckCSP, [128]],
+   [-1, 1, Conv, [256, 3, 2]],  # 3-P3/8
+   [-1, 9, BottleneckCSP, [256]],
+   [-1, 1, Conv, [512, 3, 2]],  # 5-P4/16
+   [-1, 9, BottleneckCSP, [512]],
+   [-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32
+   [-1, 1, SPP, [1024, [5, 9, 13]]],
+   [-1, 3, BottleneckCSP, [1024, False]],  # 9
+  ]
+
+# YOLOv5 head
+head:
+  [[-1, 1, Conv, [512, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 6], 1, Concat, [1]],  # cat backbone P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 13
+
+   [-1, 1, Conv, [256, 1, 1]],
+   [-1, 1, nn.Upsample, [None, 2, 'nearest']],
+   [[-1, 4], 1, Concat, [1]],  # cat backbone P3
+   [-1, 3, BottleneckCSP, [256, False]],  # 17 (P3/8-small)
+
+   [-1, 1, Conv, [256, 3, 2]],
+   [[-1, 14], 1, Concat, [1]],  # cat head P4
+   [-1, 3, BottleneckCSP, [512, False]],  # 20 (P4/16-medium)
+
+   [-1, 1, Conv, [512, 3, 2]],
+   [[-1, 10], 1, Concat, [1]],  # cat head P5
+   [-1, 3, BottleneckCSP, [1024, False]],  # 23 (P5/32-large)
+
+   [[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)
+  ]

requirements.txt

@@ -0,0 +1,10 @@
+matplotlib>=3.2.2
+numpy>=1.18.5
+opencv-python>=4.1.2
+opencv-python-headless==4.9.0.80  
+pillow
+PyYAML>=5.3
+scipy>=1.4.1
+torch>=1.6.0
+torchvision>=0.7.0
+tqdm>=4.41.0

utils/activations.py

@@ -0,0 +1,69 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+# Swish https://arxiv.org/pdf/1905.02244.pdf ---------------------------------------------------------------------------
+class Swish(nn.Module):  #
+    @staticmethod
+    def forward(x):
+        return x * torch.sigmoid(x)
+
+
+class HardSwish(nn.Module):
+    @staticmethod
+    def forward(x):
+        return x * F.hardtanh(x + 3, 0., 6., True) / 6.
+
+
+class MemoryEfficientSwish(nn.Module):
+    class F(torch.autograd.Function):
+        @staticmethod
+        def forward(ctx, x):
+            ctx.save_for_backward(x)
+            return x * torch.sigmoid(x)
+
+        @staticmethod
+        def backward(ctx, grad_output):
+            x = ctx.saved_tensors[0]
+            sx = torch.sigmoid(x)
+            return grad_output * (sx * (1 + x * (1 - sx)))
+
+    def forward(self, x):
+        return self.F.apply(x)
+
+
+# Mish https://github.com/digantamisra98/Mish --------------------------------------------------------------------------
+class Mish(nn.Module):
+    @staticmethod
+    def forward(x):
+        return x * F.softplus(x).tanh()
+
+
+class MemoryEfficientMish(nn.Module):
+    class F(torch.autograd.Function):
+        @staticmethod
+        def forward(ctx, x):
+            ctx.save_for_backward(x)
+            return x.mul(torch.tanh(F.softplus(x)))  # x * tanh(ln(1 + exp(x)))
+
+        @staticmethod
+        def backward(ctx, grad_output):
+            x = ctx.saved_tensors[0]
+            sx = torch.sigmoid(x)
+            fx = F.softplus(x).tanh()
+            return grad_output * (fx + x * sx * (1 - fx * fx))
+
+    def forward(self, x):
+        return self.F.apply(x)
+
+
+# FReLU https://arxiv.org/abs/2007.11824 -------------------------------------------------------------------------------
+class FReLU(nn.Module):
+    def __init__(self, c1, k=3):  # ch_in, kernel
+        super().__init__()
+        self.conv = nn.Conv2d(c1, c1, k, 1, 1, groups=c1)
+        self.bn = nn.BatchNorm2d(c1)
+
+    def forward(self, x):
+        return torch.max(x, self.bn(self.conv(x)))

utils/datasets.py

@@ -0,0 +1,907 @@
+import glob
+import math
+import os
+import random
+import shutil
+import time
+from pathlib import Path
+from threading import Thread
+
+import cv2
+import numpy as np
+import torch
+from PIL import Image, ExifTags
+from torch.utils.data import Dataset
+from tqdm import tqdm
+
+from utils.general import xyxy2xywh, xywh2xyxy, torch_distributed_zero_first
+
+help_url = 'https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data'
+img_formats = ['.bmp', '.jpg', '.jpeg', '.png', '.tif', '.tiff', '.dng']
+vid_formats = ['.mov', '.avi', '.mp4', '.mpg', '.mpeg', '.m4v', '.wmv', '.mkv']
+
+# Get orientation exif tag
+for orientation in ExifTags.TAGS.keys():
+    if ExifTags.TAGS[orientation] == 'Orientation':
+        break
+
+
+def get_hash(files):
+    # Returns a single hash value of a list of files
+    return sum(os.path.getsize(f) for f in files if os.path.isfile(f))
+
+
+def exif_size(img):
+    # Returns exif-corrected PIL size
+    s = img.size  # (width, height)
+    try:
+        rotation = dict(img._getexif().items())[orientation]
+        if rotation == 6:  # rotation 270
+            s = (s[1], s[0])
+        elif rotation == 8:  # rotation 90
+            s = (s[1], s[0])
+    except:
+        pass
+
+    return s
+
+
+def create_dataloader(path, imgsz, batch_size, stride, opt, hyp=None, augment=False, cache=False, pad=0.0, rect=False,
+                      local_rank=-1, world_size=1):
+    # Make sure only the first process in DDP process the dataset first, and the following others can use the cache.
+    with torch_distributed_zero_first(local_rank):
+        dataset = LoadImagesAndLabels(path, imgsz, batch_size,
+                                      augment=augment,  # augment images
+                                      hyp=hyp,  # augmentation hyperparameters
+                                      rect=rect,  # rectangular training
+                                      cache_images=cache,
+                                      single_cls=opt.single_cls,
+                                      stride=int(stride),
+                                      pad=pad)
+
+    batch_size = min(batch_size, len(dataset))
+    nw = min([os.cpu_count() // world_size, batch_size if batch_size > 1 else 0, 8])  # number of workers
+    train_sampler = torch.utils.data.distributed.DistributedSampler(dataset) if local_rank != -1 else None
+    dataloader = torch.utils.data.DataLoader(dataset,
+                                             batch_size=batch_size,
+                                             num_workers=nw,
+                                             sampler=train_sampler,
+                                             pin_memory=True,
+                                             collate_fn=LoadImagesAndLabels.collate_fn)
+    return dataloader, dataset
+
+
+class LoadImages:  # for inference
+    def __init__(self, path, img_size=640):
+        p = str(Path(path))  # os-agnostic
+        p = os.path.abspath(p)  # absolute path
+        if '*' in p:
+            files = sorted(glob.glob(p))  # glob
+        elif os.path.isdir(p):
+            files = sorted(glob.glob(os.path.join(p, '*.*')))  # dir
+        elif os.path.isfile(p):
+            files = [p]  # files
+        else:
+            raise Exception('ERROR: %s does not exist' % p)
+
+        images = [x for x in files if os.path.splitext(x)[-1].lower() in img_formats]
+        videos = [x for x in files if os.path.splitext(x)[-1].lower() in vid_formats]
+        ni, nv = len(images), len(videos)
+
+        self.img_size = img_size
+        self.files = images + videos
+        self.nf = ni + nv  # number of files
+        self.video_flag = [False] * ni + [True] * nv
+        self.mode = 'images'
+        if any(videos):
+            self.new_video(videos[0])  # new video
+        else:
+            self.cap = None
+        assert self.nf > 0, 'No images or videos found in %s. Supported formats are:\nimages: %s\nvideos: %s' % \
+                            (p, img_formats, vid_formats)
+
+    def __iter__(self):
+        self.count = 0
+        return self
+
+    def __next__(self):
+        if self.count == self.nf:
+            raise StopIteration
+        path = self.files[self.count]
+
+        if self.video_flag[self.count]:
+            # Read video
+            self.mode = 'video'
+            ret_val, img0 = self.cap.read()
+            if not ret_val:
+                self.count += 1
+                self.cap.release()
+                if self.count == self.nf:  # last video
+                    raise StopIteration
+                else:
+                    path = self.files[self.count]
+                    self.new_video(path)
+                    ret_val, img0 = self.cap.read()
+
+            self.frame += 1
+            print('video %g/%g (%g/%g) %s: ' % (self.count + 1, self.nf, self.frame, self.nframes, path), end='')
+
+        else:
+            # Read image
+            self.count += 1
+            img0 = cv2.imread(path)  # BGR
+            assert img0 is not None, 'Image Not Found ' + path
+            print('image %g/%g %s: ' % (self.count, self.nf, path), end='')
+
+        # Padded resize
+        img = letterbox(img0, new_shape=self.img_size)[0]
+
+        # Convert
+        img = img[:, :, ::-1].transpose(2, 0, 1)  # BGR to RGB, to 3x416x416
+        img = np.ascontiguousarray(img)
+
+        # cv2.imwrite(path + '.letterbox.jpg', 255 * img.transpose((1, 2, 0))[:, :, ::-1])  # save letterbox image
+        return path, img, img0, self.cap
+
+    def new_video(self, path):
+        self.frame = 0
+        self.cap = cv2.VideoCapture(path)
+        self.nframes = int(self.cap.get(cv2.CAP_PROP_FRAME_COUNT))
+
+    def __len__(self):
+        return self.nf  # number of files
+
+
+class LoadWebcam:  # for inference
+    def __init__(self, pipe=0, img_size=640):
+        self.img_size = img_size
+
+        if pipe == '0':
+            pipe = 0  # local camera
+        # pipe = 'rtsp://192.168.1.64/1'  # IP camera
+        # pipe = 'rtsp://username:[email protected]/1'  # IP camera with login
+        # pipe = 'rtsp://170.93.143.139/rtplive/470011e600ef003a004ee33696235daa'  # IP traffic camera
+        # pipe = 'http://wmccpinetop.axiscam.net/mjpg/video.mjpg'  # IP golf camera
+
+        # https://answers.opencv.org/question/215996/changing-gstreamer-pipeline-to-opencv-in-pythonsolved/
+        # pipe = '"rtspsrc location="rtsp://username:[email protected]/1" latency=10 ! appsink'  # GStreamer
+
+        # https://answers.opencv.org/question/200787/video-acceleration-gstremer-pipeline-in-videocapture/
+        # https://stackoverflow.com/questions/54095699/install-gstreamer-support-for-opencv-python-package  # install help
+        # pipe = "rtspsrc location=rtsp://root:[email protected]:554/axis-media/media.amp?videocodec=h264&resolution=3840x2160 protocols=GST_RTSP_LOWER_TRANS_TCP ! rtph264depay ! queue ! vaapih264dec ! videoconvert ! appsink"  # GStreamer
+
+        self.pipe = pipe
+        self.cap = cv2.VideoCapture(pipe)  # video capture object
+        self.cap.set(cv2.CAP_PROP_BUFFERSIZE, 3)  # set buffer size
+
+    def __iter__(self):
+        self.count = -1
+        return self
+
+    def __next__(self):
+        self.count += 1
+        if cv2.waitKey(1) == ord('q'):  # q to quit
+            self.cap.release()
+            cv2.destroyAllWindows()
+            raise StopIteration
+
+        # Read frame
+        if self.pipe == 0:  # local camera
+            ret_val, img0 = self.cap.read()
+            img0 = cv2.flip(img0, 1)  # flip left-right
+        else:  # IP camera
+            n = 0
+            while True:
+                n += 1
+                self.cap.grab()
+                if n % 30 == 0:  # skip frames
+                    ret_val, img0 = self.cap.retrieve()
+                    if ret_val:
+                        break
+
+        # Print
+        assert ret_val, 'Camera Error %s' % self.pipe
+        img_path = 'webcam.jpg'
+        print('webcam %g: ' % self.count, end='')
+
+        # Padded resize
+        img = letterbox(img0, new_shape=self.img_size)[0]
+
+        # Convert
+        img = img[:, :, ::-1].transpose(2, 0, 1)  # BGR to RGB, to 3x416x416
+        img = np.ascontiguousarray(img)
+
+        return img_path, img, img0, None
+
+    def __len__(self):
+        return 0
+
+
+class LoadStreams:  # multiple IP or RTSP cameras
+    def __init__(self, sources='streams.txt', img_size=640):
+        self.mode = 'images'
+        self.img_size = img_size
+
+        if os.path.isfile(sources):
+            with open(sources, 'r') as f:
+                sources = [x.strip() for x in f.read().splitlines() if len(x.strip())]
+        else:
+            sources = [sources]
+
+        n = len(sources)
+        self.imgs = [None] * n
+        self.sources = sources
+        for i, s in enumerate(sources):
+            # Start the thread to read frames from the video stream
+            print('%g/%g: %s... ' % (i + 1, n, s), end='')
+            cap = cv2.VideoCapture(0 if s == '0' else s)
+            assert cap.isOpened(), 'Failed to open %s' % s
+            w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
+            h = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
+            fps = cap.get(cv2.CAP_PROP_FPS) % 100
+            _, self.imgs[i] = cap.read()  # guarantee first frame
+            thread = Thread(target=self.update, args=([i, cap]), daemon=True)
+            print(' success (%gx%g at %.2f FPS).' % (w, h, fps))
+            thread.start()
+        print('')  # newline
+
+        # check for common shapes
+        s = np.stack([letterbox(x, new_shape=self.img_size)[0].shape for x in self.imgs], 0)  # inference shapes
+        self.rect = np.unique(s, axis=0).shape[0] == 1  # rect inference if all shapes equal
+        if not self.rect:
+            print('WARNING: Different stream shapes detected. For optimal performance supply similarly-shaped streams.')
+
+    def update(self, index, cap):
+        # Read next stream frame in a daemon thread
+        n = 0
+        while cap.isOpened():
+            n += 1
+            # _, self.imgs[index] = cap.read()
+            cap.grab()
+            if n == 4:  # read every 4th frame
+                _, self.imgs[index] = cap.retrieve()
+                n = 0
+            time.sleep(0.01)  # wait time
+
+    def __iter__(self):
+        self.count = -1
+        return self
+
+    def __next__(self):
+        self.count += 1
+        img0 = self.imgs.copy()
+        if cv2.waitKey(1) == ord('q'):  # q to quit
+            cv2.destroyAllWindows()
+            raise StopIteration
+
+        # Letterbox
+        img = [letterbox(x, new_shape=self.img_size, auto=self.rect)[0] for x in img0]
+
+        # Stack
+        img = np.stack(img, 0)
+
+        # Convert
+        img = img[:, :, :, ::-1].transpose(0, 3, 1, 2)  # BGR to RGB, to bsx3x416x416
+        img = np.ascontiguousarray(img)
+
+        return self.sources, img, img0, None
+
+    def __len__(self):
+        return 0  # 1E12 frames = 32 streams at 30 FPS for 30 years
+
+
+class LoadImagesAndLabels(Dataset):  # for training/testing
+    def __init__(self, path, img_size=640, batch_size=16, augment=False, hyp=None, rect=False, image_weights=False,
+                 cache_images=False, single_cls=False, stride=32, pad=0.0):
+        try:
+            f = []  # image files
+            for p in path if isinstance(path, list) else [path]:
+                p = str(Path(p))  # os-agnostic
+                parent = str(Path(p).parent) + os.sep
+                if os.path.isfile(p):  # file
+                    with open(p, 'r') as t:
+                        t = t.read().splitlines()
+                        f += [x.replace('./', parent) if x.startswith('./') else x for x in t]  # local to global path
+                elif os.path.isdir(p):  # folder
+                    f += glob.iglob(p + os.sep + '*.*')
+                else:
+                    raise Exception('%s does not exist' % p)
+            self.img_files = sorted(
+                [x.replace('/', os.sep) for x in f if os.path.splitext(x)[-1].lower() in img_formats])
+        except Exception as e:
+            raise Exception('Error loading data from %s: %s\nSee %s' % (path, e, help_url))
+
+        n = len(self.img_files)
+        assert n > 0, 'No images found in %s. See %s' % (path, help_url)
+        bi = np.floor(np.arange(n) / batch_size).astype(np.int)  # batch index
+        nb = bi[-1] + 1  # number of batches
+
+        self.n = n  # number of images
+        self.batch = bi  # batch index of image
+        self.img_size = img_size
+        self.augment = augment
+        self.hyp = hyp
+        self.image_weights = image_weights
+        self.rect = False if image_weights else rect
+        self.mosaic = self.augment and not self.rect  # load 4 images at a time into a mosaic (only during training)
+        self.mosaic_border = [-img_size // 2, -img_size // 2]
+        self.stride = stride
+
+        # Define labels
+        self.label_files = [x.replace('images', 'labels').replace(os.path.splitext(x)[-1], '.txt') for x in
+                            self.img_files]
+
+        # Check cache
+        cache_path = str(Path(self.label_files[0]).parent) + '.cache'  # cached labels
+        if os.path.isfile(cache_path):
+            cache = torch.load(cache_path)  # load
+            if cache['hash'] != get_hash(self.label_files + self.img_files):  # dataset changed
+                cache = self.cache_labels(cache_path)  # re-cache
+        else:
+            cache = self.cache_labels(cache_path)  # cache
+
+        # Get labels
+        labels, shapes = zip(*[cache[x] for x in self.img_files])
+        self.shapes = np.array(shapes, dtype=np.float64)
+        self.labels = list(labels)
+
+        # Rectangular Training  https://github.com/ultralytics/yolov3/issues/232
+        if self.rect:
+            # Sort by aspect ratio
+            s = self.shapes  # wh
+            ar = s[:, 1] / s[:, 0]  # aspect ratio
+            irect = ar.argsort()
+            self.img_files = [self.img_files[i] for i in irect]
+            self.label_files = [self.label_files[i] for i in irect]
+            self.labels = [self.labels[i] for i in irect]
+            self.shapes = s[irect]  # wh
+            ar = ar[irect]
+
+            # Set training image shapes
+            shapes = [[1, 1]] * nb
+            for i in range(nb):
+                ari = ar[bi == i]
+                mini, maxi = ari.min(), ari.max()
+                if maxi < 1:
+                    shapes[i] = [maxi, 1]
+                elif mini > 1:
+                    shapes[i] = [1, 1 / mini]
+
+            self.batch_shapes = np.ceil(np.array(shapes) * img_size / stride + pad).astype(np.int) * stride
+
+        # Cache labels
+        create_datasubset, extract_bounding_boxes, labels_loaded = False, False, False
+        nm, nf, ne, ns, nd = 0, 0, 0, 0, 0  # number missing, found, empty, datasubset, duplicate
+        pbar = tqdm(self.label_files)
+        for i, file in enumerate(pbar):
+            l = self.labels[i]  # label
+            if l.shape[0]:
+                assert l.shape[1] == 5, '> 5 label columns: %s' % file
+                assert (l >= 0).all(), 'negative labels: %s' % file
+                assert (l[:, 1:] <= 1).all(), 'non-normalized or out of bounds coordinate labels: %s' % file
+                if np.unique(l, axis=0).shape[0] < l.shape[0]:  # duplicate rows
+                    nd += 1  # print('WARNING: duplicate rows in %s' % self.label_files[i])  # duplicate rows
+                if single_cls:
+                    l[:, 0] = 0  # force dataset into single-class mode
+                self.labels[i] = l
+                nf += 1  # file found
+
+                # Create subdataset (a smaller dataset)
+                if create_datasubset and ns < 1E4:
+                    if ns == 0:
+                        create_folder(path='./datasubset')
+                        os.makedirs('./datasubset/images')
+                    exclude_classes = 43
+                    if exclude_classes not in l[:, 0]:
+                        ns += 1
+                        # shutil.copy(src=self.img_files[i], dst='./datasubset/images/')  # copy image
+                        with open('./datasubset/images.txt', 'a') as f:
+                            f.write(self.img_files[i] + '\n')
+
+                # Extract object detection boxes for a second stage classifier
+                if extract_bounding_boxes:
+                    p = Path(self.img_files[i])
+                    img = cv2.imread(str(p))
+                    h, w = img.shape[:2]
+                    for j, x in enumerate(l):
+                        f = '%s%sclassifier%s%g_%g_%s' % (p.parent.parent, os.sep, os.sep, x[0], j, p.name)
+                        if not os.path.exists(Path(f).parent):
+                            os.makedirs(Path(f).parent)  # make new output folder
+
+                        b = x[1:] * [w, h, w, h]  # box
+                        b[2:] = b[2:].max()  # rectangle to square
+                        b[2:] = b[2:] * 1.3 + 30  # pad
+                        b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(np.int)
+
+                        b[[0, 2]] = np.clip(b[[0, 2]], 0, w)  # clip boxes outside of image
+                        b[[1, 3]] = np.clip(b[[1, 3]], 0, h)
+                        assert cv2.imwrite(f, img[b[1]:b[3], b[0]:b[2]]), 'Failure extracting classifier boxes'
+            else:
+                ne += 1  # print('empty labels for image %s' % self.img_files[i])  # file empty
+                # os.system("rm '%s' '%s'" % (self.img_files[i], self.label_files[i]))  # remove
+
+            pbar.desc = 'Scanning labels %s (%g found, %g missing, %g empty, %g duplicate, for %g images)' % (
+                cache_path, nf, nm, ne, nd, n)
+        if nf == 0:
+            s = 'WARNING: No labels found in %s. See %s' % (os.path.dirname(file) + os.sep, help_url)
+            print(s)
+            assert not augment, '%s. Can not train without labels.' % s
+
+        # Cache images into memory for faster training (WARNING: large datasets may exceed system RAM)
+        self.imgs = [None] * n
+        if cache_images:
+            gb = 0  # Gigabytes of cached images
+            pbar = tqdm(range(len(self.img_files)), desc='Caching images')
+            self.img_hw0, self.img_hw = [None] * n, [None] * n
+            for i in pbar:  # max 10k images
+                self.imgs[i], self.img_hw0[i], self.img_hw[i] = load_image(self, i)  # img, hw_original, hw_resized
+                gb += self.imgs[i].nbytes
+                pbar.desc = 'Caching images (%.1fGB)' % (gb / 1E9)
+
+    def cache_labels(self, path='labels.cache'):
+        # Cache dataset labels, check images and read shapes
+        x = {}  # dict
+        pbar = tqdm(zip(self.img_files, self.label_files), desc='Scanning images', total=len(self.img_files))
+        for (img, label) in pbar:
+            try:
+                l = []
+                image = Image.open(img)
+                image.verify()  # PIL verify
+                # _ = io.imread(img)  # skimage verify (from skimage import io)
+                shape = exif_size(image)  # image size
+                assert (shape[0] > 9) & (shape[1] > 9), 'image size <10 pixels'
+                if os.path.isfile(label):
+                    with open(label, 'r') as f:
+                        l = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32)  # labels
+                if len(l) == 0:
+                    l = np.zeros((0, 5), dtype=np.float32)
+                x[img] = [l, shape]
+            except Exception as e:
+                x[img] = [None, None]
+                print('WARNING: %s: %s' % (img, e))
+
+        x['hash'] = get_hash(self.label_files + self.img_files)
+        torch.save(x, path)  # save for next time
+        return x
+
+    def __len__(self):
+        return len(self.img_files)
+
+    # def __iter__(self):
+    #     self.count = -1
+    #     print('ran dataset iter')
+    #     #self.shuffled_vector = np.random.permutation(self.nF) if self.augment else np.arange(self.nF)
+    #     return self
+
+    def __getitem__(self, index):
+        if self.image_weights:
+            index = self.indices[index]
+
+        hyp = self.hyp
+        if self.mosaic:
+            # Load mosaic
+            img, labels = load_mosaic(self, index)
+            shapes = None
+
+            # MixUp https://arxiv.org/pdf/1710.09412.pdf
+            if random.random() < hyp['mixup']:
+                img2, labels2 = load_mosaic(self, random.randint(0, len(self.labels) - 1))
+                r = np.random.beta(8.0, 8.0)  # mixup ratio, alpha=beta=8.0
+                img = (img * r + img2 * (1 - r)).astype(np.uint8)
+                labels = np.concatenate((labels, labels2), 0)
+
+        else:
+            # Load image
+            img, (h0, w0), (h, w) = load_image(self, index)
+
+            # Letterbox
+            shape = self.batch_shapes[self.batch[index]] if self.rect else self.img_size  # final letterboxed shape
+            img, ratio, pad = letterbox(img, shape, auto=False, scaleup=self.augment)
+            shapes = (h0, w0), ((h / h0, w / w0), pad)  # for COCO mAP rescaling
+
+            # Load labels
+            labels = []
+            x = self.labels[index]
+            if x.size > 0:
+                # Normalized xywh to pixel xyxy format
+                labels = x.copy()
+                labels[:, 1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + pad[0]  # pad width
+                labels[:, 2] = ratio[1] * h * (x[:, 2] - x[:, 4] / 2) + pad[1]  # pad height
+                labels[:, 3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + pad[0]
+                labels[:, 4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + pad[1]
+
+        if self.augment:
+            # Augment imagespace
+            if not self.mosaic:
+                img, labels = random_perspective(img, labels,
+                                                 degrees=hyp['degrees'],
+                                                 translate=hyp['translate'],
+                                                 scale=hyp['scale'],
+                                                 shear=hyp['shear'],
+                                                 perspective=hyp['perspective'])
+
+            # Augment colorspace
+            augment_hsv(img, hgain=hyp['hsv_h'], sgain=hyp['hsv_s'], vgain=hyp['hsv_v'])
+
+            # Apply cutouts
+            # if random.random() < 0.9:
+            #     labels = cutout(img, labels)
+
+        nL = len(labels)  # number of labels
+        if nL:
+            labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])  # convert xyxy to xywh
+            labels[:, [2, 4]] /= img.shape[0]  # normalized height 0-1
+            labels[:, [1, 3]] /= img.shape[1]  # normalized width 0-1
+
+        if self.augment:
+            # flip up-down
+            if random.random() < hyp['flipud']:
+                img = np.flipud(img)
+                if nL:
+                    labels[:, 2] = 1 - labels[:, 2]
+
+            # flip left-right
+            if random.random() < hyp['fliplr']:
+                img = np.fliplr(img)
+                if nL:
+                    labels[:, 1] = 1 - labels[:, 1]
+
+        labels_out = torch.zeros((nL, 6))
+        if nL:
+            labels_out[:, 1:] = torch.from_numpy(labels)
+
+        # Convert
+        img = img[:, :, ::-1].transpose(2, 0, 1)  # BGR to RGB, to 3x416x416
+        img = np.ascontiguousarray(img)
+
+        return torch.from_numpy(img), labels_out, self.img_files[index], shapes
+
+    @staticmethod
+    def collate_fn(batch):
+        img, label, path, shapes = zip(*batch)  # transposed
+        for i, l in enumerate(label):
+            l[:, 0] = i  # add target image index for build_targets()
+        return torch.stack(img, 0), torch.cat(label, 0), path, shapes
+
+
+# Ancillary functions --------------------------------------------------------------------------------------------------
+def load_image(self, index):
+    # loads 1 image from dataset, returns img, original hw, resized hw
+    img = self.imgs[index]
+    if img is None:  # not cached
+        path = self.img_files[index]
+        img = cv2.imread(path)  # BGR
+        assert img is not None, 'Image Not Found ' + path
+        h0, w0 = img.shape[:2]  # orig hw
+        r = self.img_size / max(h0, w0)  # resize image to img_size
+        if r != 1:  # always resize down, only resize up if training with augmentation
+            interp = cv2.INTER_AREA if r < 1 and not self.augment else cv2.INTER_LINEAR
+            img = cv2.resize(img, (int(w0 * r), int(h0 * r)), interpolation=interp)
+        return img, (h0, w0), img.shape[:2]  # img, hw_original, hw_resized
+    else:
+        return self.imgs[index], self.img_hw0[index], self.img_hw[index]  # img, hw_original, hw_resized
+
+
+def augment_hsv(img, hgain=0.5, sgain=0.5, vgain=0.5):
+    r = np.random.uniform(-1, 1, 3) * [hgain, sgain, vgain] + 1  # random gains
+    hue, sat, val = cv2.split(cv2.cvtColor(img, cv2.COLOR_BGR2HSV))
+    dtype = img.dtype  # uint8
+
+    x = np.arange(0, 256, dtype=np.int16)
+    lut_hue = ((x * r[0]) % 180).astype(dtype)
+    lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
+    lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
+
+    img_hsv = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val))).astype(dtype)
+    cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)  # no return needed
+
+    # Histogram equalization
+    # if random.random() < 0.2:
+    #     for i in range(3):
+    #         img[:, :, i] = cv2.equalizeHist(img[:, :, i])
+
+
+def load_mosaic(self, index):
+    # loads images in a mosaic
+
+    labels4 = []
+    s = self.img_size
+    yc, xc = s, s  # mosaic center x, y
+    indices = [index] + [random.randint(0, len(self.labels) - 1) for _ in range(3)]  # 3 additional image indices
+    for i, index in enumerate(indices):
+        # Load image
+        img, _, (h, w) = load_image(self, index)
+
+        # place img in img4
+        if i == 0:  # top left
+            img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8)  # base image with 4 tiles
+            x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc  # xmin, ymin, xmax, ymax (large image)
+            x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h  # xmin, ymin, xmax, ymax (small image)
+        elif i == 1:  # top right
+            x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
+            x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
+        elif i == 2:  # bottom left
+            x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
+            x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, max(xc, w), min(y2a - y1a, h)
+        elif i == 3:  # bottom right
+            x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
+            x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
+
+        img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax]
+        padw = x1a - x1b
+        padh = y1a - y1b
+
+        # Labels
+        x = self.labels[index]
+        labels = x.copy()
+        if x.size > 0:  # Normalized xywh to pixel xyxy format
+            labels[:, 1] = w * (x[:, 1] - x[:, 3] / 2) + padw
+            labels[:, 2] = h * (x[:, 2] - x[:, 4] / 2) + padh
+            labels[:, 3] = w * (x[:, 1] + x[:, 3] / 2) + padw
+            labels[:, 4] = h * (x[:, 2] + x[:, 4] / 2) + padh
+        labels4.append(labels)
+
+    # Concat/clip labels
+    if len(labels4):
+        labels4 = np.concatenate(labels4, 0)
+        # np.clip(labels4[:, 1:] - s / 2, 0, s, out=labels4[:, 1:])  # use with center crop
+        np.clip(labels4[:, 1:], 0, 2 * s, out=labels4[:, 1:])  # use with random_affine
+
+        # Replicate
+        # img4, labels4 = replicate(img4, labels4)
+
+    # Augment
+    # img4 = img4[s // 2: int(s * 1.5), s // 2:int(s * 1.5)]  # center crop (WARNING, requires box pruning)
+    img4, labels4 = random_perspective(img4, labels4,
+                                       degrees=self.hyp['degrees'],
+                                       translate=self.hyp['translate'],
+                                       scale=self.hyp['scale'],
+                                       shear=self.hyp['shear'],
+                                       perspective=self.hyp['perspective'],
+                                       border=self.mosaic_border)  # border to remove
+
+    return img4, labels4
+
+
+def replicate(img, labels):
+    # Replicate labels
+    h, w = img.shape[:2]
+    boxes = labels[:, 1:].astype(int)
+    x1, y1, x2, y2 = boxes.T
+    s = ((x2 - x1) + (y2 - y1)) / 2  # side length (pixels)
+    for i in s.argsort()[:round(s.size * 0.5)]:  # smallest indices
+        x1b, y1b, x2b, y2b = boxes[i]
+        bh, bw = y2b - y1b, x2b - x1b
+        yc, xc = int(random.uniform(0, h - bh)), int(random.uniform(0, w - bw))  # offset x, y
+        x1a, y1a, x2a, y2a = [xc, yc, xc + bw, yc + bh]
+        img[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b]  # img4[ymin:ymax, xmin:xmax]
+        labels = np.append(labels, [[labels[i, 0], x1a, y1a, x2a, y2a]], axis=0)
+
+    return img, labels
+
+
+def letterbox(img, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True):
+    # Resize image to a 32-pixel-multiple rectangle https://github.com/ultralytics/yolov3/issues/232
+    shape = img.shape[:2]  # current shape [height, width]
+    if isinstance(new_shape, int):
+        new_shape = (new_shape, new_shape)
+
+    # Scale ratio (new / old)
+    r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
+    if not scaleup:  # only scale down, do not scale up (for better test mAP)
+        r = min(r, 1.0)
+
+    # Compute padding
+    ratio = r, r  # width, height ratios
+    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
+    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding
+    if auto:  # minimum rectangle
+        dw, dh = np.mod(dw, 64), np.mod(dh, 64)  # wh padding
+    elif scaleFill:  # stretch
+        dw, dh = 0.0, 0.0
+        new_unpad = (new_shape[1], new_shape[0])
+        ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios
+
+    dw /= 2  # divide padding into 2 sides
+    dh /= 2
+
+    if shape[::-1] != new_unpad:  # resize
+        img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR)
+    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
+    left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
+    img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border
+    return img, ratio, (dw, dh)
+
+
+def random_perspective(img, targets=(), degrees=10, translate=.1, scale=.1, shear=10, perspective=0.0, border=(0, 0)):
+    # torchvision.transforms.RandomAffine(degrees=(-10, 10), translate=(.1, .1), scale=(.9, 1.1), shear=(-10, 10))
+    # targets = [cls, xyxy]
+
+    height = img.shape[0] + border[0] * 2  # shape(h,w,c)
+    width = img.shape[1] + border[1] * 2
+
+    # Center
+    C = np.eye(3)
+    C[0, 2] = -img.shape[1] / 2  # x translation (pixels)
+    C[1, 2] = -img.shape[0] / 2  # y translation (pixels)
+
+    # Perspective
+    P = np.eye(3)
+    P[2, 0] = random.uniform(-perspective, perspective)  # x perspective (about y)
+    P[2, 1] = random.uniform(-perspective, perspective)  # y perspective (about x)
+
+    # Rotation and Scale
+    R = np.eye(3)
+    a = random.uniform(-degrees, degrees)
+    # a += random.choice([-180, -90, 0, 90])  # add 90deg rotations to small rotations
+    s = random.uniform(1 - scale, 1 + scale)
+    # s = 2 ** random.uniform(-scale, scale)
+    R[:2] = cv2.getRotationMatrix2D(angle=a, center=(0, 0), scale=s)
+
+    # Shear
+    S = np.eye(3)
+    S[0, 1] = math.tan(random.uniform(-shear, shear) * math.pi / 180)  # x shear (deg)
+    S[1, 0] = math.tan(random.uniform(-shear, shear) * math.pi / 180)  # y shear (deg)
+
+    # Translation
+    T = np.eye(3)
+    T[0, 2] = random.uniform(0.5 - translate, 0.5 + translate) * width  # x translation (pixels)
+    T[1, 2] = random.uniform(0.5 - translate, 0.5 + translate) * height  # y translation (pixels)
+
+    # Combined rotation matrix
+    M = T @ S @ R @ P @ C  # order of operations (right to left) is IMPORTANT
+    if (border[0] != 0) or (border[1] != 0) or (M != np.eye(3)).any():  # image changed
+        if perspective:
+            img = cv2.warpPerspective(img, M, dsize=(width, height), borderValue=(114, 114, 114))
+        else:  # affine
+            img = cv2.warpAffine(img, M[:2], dsize=(width, height), borderValue=(114, 114, 114))
+
+    # Visualize
+    # import matplotlib.pyplot as plt
+    # ax = plt.subplots(1, 2, figsize=(12, 6))[1].ravel()
+    # ax[0].imshow(img[:, :, ::-1])  # base
+    # ax[1].imshow(img2[:, :, ::-1])  # warped
+
+    # Transform label coordinates
+    n = len(targets)
+    if n:
+        # warp points
+        xy = np.ones((n * 4, 3))
+        xy[:, :2] = targets[:, [1, 2, 3, 4, 1, 4, 3, 2]].reshape(n * 4, 2)  # x1y1, x2y2, x1y2, x2y1
+        xy = xy @ M.T  # transform
+        if perspective:
+            xy = (xy[:, :2] / xy[:, 2:3]).reshape(n, 8)  # rescale
+        else:  # affine
+            xy = xy[:, :2].reshape(n, 8)
+
+        # create new boxes
+        x = xy[:, [0, 2, 4, 6]]
+        y = xy[:, [1, 3, 5, 7]]
+        xy = np.concatenate((x.min(1), y.min(1), x.max(1), y.max(1))).reshape(4, n).T
+
+        # # apply angle-based reduction of bounding boxes
+        # radians = a * math.pi / 180
+        # reduction = max(abs(math.sin(radians)), abs(math.cos(radians))) ** 0.5
+        # x = (xy[:, 2] + xy[:, 0]) / 2
+        # y = (xy[:, 3] + xy[:, 1]) / 2
+        # w = (xy[:, 2] - xy[:, 0]) * reduction
+        # h = (xy[:, 3] - xy[:, 1]) * reduction
+        # xy = np.concatenate((x - w / 2, y - h / 2, x + w / 2, y + h / 2)).reshape(4, n).T
+
+        # clip boxes
+        xy[:, [0, 2]] = xy[:, [0, 2]].clip(0, width)
+        xy[:, [1, 3]] = xy[:, [1, 3]].clip(0, height)
+
+        # filter candidates
+        i = box_candidates(box1=targets[:, 1:5].T * s, box2=xy.T)
+        targets = targets[i]
+        targets[:, 1:5] = xy[i]
+
+    return img, targets
+
+
+def box_candidates(box1, box2, wh_thr=2, ar_thr=20, area_thr=0.2):  # box1(4,n), box2(4,n)
+    # Compute candidate boxes: box1 before augment, box2 after augment, wh_thr (pixels), aspect_ratio_thr, area_ratio
+    w1, h1 = box1[2] - box1[0], box1[3] - box1[1]
+    w2, h2 = box2[2] - box2[0], box2[3] - box2[1]
+    ar = np.maximum(w2 / (h2 + 1e-16), h2 / (w2 + 1e-16))  # aspect ratio
+    return (w2 > wh_thr) & (h2 > wh_thr) & (w2 * h2 / (w1 * h1 + 1e-16) > area_thr) & (ar < ar_thr)  # candidates
+
+
+def cutout(image, labels):
+    # Applies image cutout augmentation https://arxiv.org/abs/1708.04552
+    h, w = image.shape[:2]
+
+    def bbox_ioa(box1, box2):
+        # Returns the intersection over box2 area given box1, box2. box1 is 4, box2 is nx4. boxes are x1y1x2y2
+        box2 = box2.transpose()
+
+        # Get the coordinates of bounding boxes
+        b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
+        b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
+
+        # Intersection area
+        inter_area = (np.minimum(b1_x2, b2_x2) - np.maximum(b1_x1, b2_x1)).clip(0) * \
+                     (np.minimum(b1_y2, b2_y2) - np.maximum(b1_y1, b2_y1)).clip(0)
+
+        # box2 area
+        box2_area = (b2_x2 - b2_x1) * (b2_y2 - b2_y1) + 1e-16
+
+        # Intersection over box2 area
+        return inter_area / box2_area
+
+    # create random masks
+    scales = [0.5] * 1 + [0.25] * 2 + [0.125] * 4 + [0.0625] * 8 + [0.03125] * 16  # image size fraction
+    for s in scales:
+        mask_h = random.randint(1, int(h * s))
+        mask_w = random.randint(1, int(w * s))
+
+        # box
+        xmin = max(0, random.randint(0, w) - mask_w // 2)
+        ymin = max(0, random.randint(0, h) - mask_h // 2)
+        xmax = min(w, xmin + mask_w)
+        ymax = min(h, ymin + mask_h)
+
+        # apply random color mask
+        image[ymin:ymax, xmin:xmax] = [random.randint(64, 191) for _ in range(3)]
+
+        # return unobscured labels
+        if len(labels) and s > 0.03:
+            box = np.array([xmin, ymin, xmax, ymax], dtype=np.float32)
+            ioa = bbox_ioa(box, labels[:, 1:5])  # intersection over area
+            labels = labels[ioa < 0.60]  # remove >60% obscured labels
+
+    return labels
+
+
+def reduce_img_size(path='path/images', img_size=1024):  # from utils.datasets import *; reduce_img_size()
+    # creates a new ./images_reduced folder with reduced size images of maximum size img_size
+    path_new = path + '_reduced'  # reduced images path
+    create_folder(path_new)
+    for f in tqdm(glob.glob('%s/*.*' % path)):
+        try:
+            img = cv2.imread(f)
+            h, w = img.shape[:2]
+            r = img_size / max(h, w)  # size ratio
+            if r < 1.0:
+                img = cv2.resize(img, (int(w * r), int(h * r)), interpolation=cv2.INTER_AREA)  # _LINEAR fastest
+            fnew = f.replace(path, path_new)  # .replace(Path(f).suffix, '.jpg')
+            cv2.imwrite(fnew, img)
+        except:
+            print('WARNING: image failure %s' % f)
+
+
+def recursive_dataset2bmp(dataset='path/dataset_bmp'):  # from utils.datasets import *; recursive_dataset2bmp()
+    # Converts dataset to bmp (for faster training)
+    formats = [x.lower() for x in img_formats] + [x.upper() for x in img_formats]
+    for a, b, files in os.walk(dataset):
+        for file in tqdm(files, desc=a):
+            p = a + '/' + file
+            s = Path(file).suffix
+            if s == '.txt':  # replace text
+                with open(p, 'r') as f:
+                    lines = f.read()
+                for f in formats:
+                    lines = lines.replace(f, '.bmp')
+                with open(p, 'w') as f:
+                    f.write(lines)
+            elif s in formats:  # replace image
+                cv2.imwrite(p.replace(s, '.bmp'), cv2.imread(p))
+                if s != '.bmp':
+                    os.system("rm '%s'" % p)
+
+
+def imagelist2folder(path='path/images.txt'):  # from utils.datasets import *; imagelist2folder()
+    # Copies all the images in a text file (list of images) into a folder
+    create_folder(path[:-4])
+    with open(path, 'r') as f:
+        for line in f.read().splitlines():
+            os.system('cp "%s" %s' % (line, path[:-4]))
+            print(line)
+
+
+def create_folder(path='./new'):
+    # Create folder
+    if os.path.exists(path):
+        shutil.rmtree(path)  # delete output folder
+    os.makedirs(path)  # make new output folder

utils/general.py

@@ -0,0 +1,1263 @@
+import glob
+import math
+import os
+import random
+import shutil
+import subprocess
+import time
+from contextlib import contextmanager
+from copy import copy
+from pathlib import Path
+from sys import platform
+
+import cv2
+import matplotlib
+import matplotlib.pyplot as plt
+import numpy as np
+import torch
+import torch.nn as nn
+import torchvision
+import yaml
+from scipy.cluster.vq import kmeans
+from scipy.signal import butter, filtfilt
+from tqdm import tqdm
+
+from utils.torch_utils import init_seeds, is_parallel
+
+# Set printoptions
+torch.set_printoptions(linewidth=320, precision=5, profile='long')
+np.set_printoptions(linewidth=320, formatter={'float_kind': '{:11.5g}'.format})  # format short g, %precision=5
+matplotlib.rc('font', **{'size': 11})
+
+# Prevent OpenCV from multithreading (to use PyTorch DataLoader)
+cv2.setNumThreads(0)
+
+
+@contextmanager
+def torch_distributed_zero_first(local_rank: int):
+    """
+    Decorator to make all processes in distributed training wait for each local_master to do something.
+    """
+    if local_rank not in [-1, 0]:
+        torch.distributed.barrier()
+    yield
+    if local_rank == 0:
+        torch.distributed.barrier()
+
+
+def init_seeds(seed=0):
+    random.seed(seed)
+    np.random.seed(seed)
+    init_seeds(seed=seed)
+
+
+def get_latest_run(search_dir='./runs'):
+    # Return path to most recent 'last.pt' in /runs (i.e. to --resume from)
+    last_list = glob.glob(f'{search_dir}/**/last*.pt', recursive=True)
+    return max(last_list, key=os.path.getctime)
+
+
+def check_git_status():
+    # Suggest 'git pull' if repo is out of date
+    if platform in ['linux', 'darwin'] and not os.path.isfile('/.dockerenv'):
+        s = subprocess.check_output('if [ -d .git ]; then git fetch && git status -uno; fi', shell=True).decode('utf-8')
+        if 'Your branch is behind' in s:
+            print(s[s.find('Your branch is behind'):s.find('\n\n')] + '\n')
+
+
+def check_img_size(img_size, s=32):
+    # Verify img_size is a multiple of stride s
+    new_size = make_divisible(img_size, int(s))  # ceil gs-multiple
+    if new_size != img_size:
+        print('WARNING: --img-size %g must be multiple of max stride %g, updating to %g' % (img_size, s, new_size))
+    return new_size
+
+
+def check_anchors(dataset, model, thr=4.0, imgsz=640):
+    # Check anchor fit to data, recompute if necessary
+    print('\nAnalyzing anchors... ', end='')
+    m = model.module.model[-1] if hasattr(model, 'module') else model.model[-1]  # Detect()
+    shapes = imgsz * dataset.shapes / dataset.shapes.max(1, keepdims=True)
+    scale = np.random.uniform(0.9, 1.1, size=(shapes.shape[0], 1))  # augment scale
+    wh = torch.tensor(np.concatenate([l[:, 3:5] * s for s, l in zip(shapes * scale, dataset.labels)])).float()  # wh
+
+    def metric(k):  # compute metric
+        r = wh[:, None] / k[None]
+        x = torch.min(r, 1. / r).min(2)[0]  # ratio metric
+        best = x.max(1)[0]  # best_x
+        aat = (x > 1. / thr).float().sum(1).mean()  # anchors above threshold
+        bpr = (best > 1. / thr).float().mean()  # best possible recall
+        return bpr, aat
+
+    bpr, aat = metric(m.anchor_grid.clone().cpu().view(-1, 2))
+    print('anchors/target = %.2f, Best Possible Recall (BPR) = %.4f' % (aat, bpr), end='')
+    if bpr < 0.98:  # threshold to recompute
+        print('. Attempting to generate improved anchors, please wait...' % bpr)
+        na = m.anchor_grid.numel() // 2  # number of anchors
+        new_anchors = kmean_anchors(dataset, n=na, img_size=imgsz, thr=thr, gen=1000, verbose=False)
+        new_bpr = metric(new_anchors.reshape(-1, 2))[0]
+        if new_bpr > bpr:  # replace anchors
+            new_anchors = torch.tensor(new_anchors, device=m.anchors.device).type_as(m.anchors)
+            m.anchor_grid[:] = new_anchors.clone().view_as(m.anchor_grid)  # for inference
+            m.anchors[:] = new_anchors.clone().view_as(m.anchors) / m.stride.to(m.anchors.device).view(-1, 1, 1)  # loss
+            check_anchor_order(m)
+            print('New anchors saved to model. Update model *.yaml to use these anchors in the future.')
+        else:
+            print('Original anchors better than new anchors. Proceeding with original anchors.')
+    print('')  # newline
+
+
+def check_anchor_order(m):
+    # Check anchor order against stride order for YOLOv5 Detect() module m, and correct if necessary
+    a = m.anchor_grid.prod(-1).view(-1)  # anchor area
+    da = a[-1] - a[0]  # delta a
+    ds = m.stride[-1] - m.stride[0]  # delta s
+    if da.sign() != ds.sign():  # same order
+        print('Reversing anchor order')
+        m.anchors[:] = m.anchors.flip(0)
+        m.anchor_grid[:] = m.anchor_grid.flip(0)
+
+
+def check_file(file):
+    # Searches for file if not found locally
+    if os.path.isfile(file) or file == '':
+        return file
+    else:
+        files = glob.glob('./**/' + file, recursive=True)  # find file
+        assert len(files), 'File Not Found: %s' % file  # assert file was found
+        return files[0]  # return first file if multiple found
+
+
+def check_dataset(dict):
+    # Download dataset if not found
+    train, val = os.path.abspath(dict['train']), os.path.abspath(dict['val'])  # data paths
+    if not (os.path.exists(train) and os.path.exists(val)):
+        print('\nWARNING: Dataset not found, nonexistant paths: %s' % [train, val])
+        if 'download' in dict:
+            s = dict['download']
+            print('Attempting autodownload from: %s' % s)
+            if s.startswith('http') and s.endswith('.zip'):  # URL
+                f = Path(s).name  # filename
+                torch.hub.download_url_to_file(s, f)
+                r = os.system('unzip -q %s -d ../ && rm %s' % (f, f))
+            else:  # bash script
+                r = os.system(s)
+            print('Dataset autodownload %s\n' % ('success' if r == 0 else 'failure'))  # analyze return value
+        else:
+            Exception('Dataset autodownload unavailable.')
+
+
+def make_divisible(x, divisor):
+    # Returns x evenly divisble by divisor
+    return math.ceil(x / divisor) * divisor
+
+
+def labels_to_class_weights(labels, nc=80):
+    # Get class weights (inverse frequency) from training labels
+    if labels[0] is None:  # no labels loaded
+        return torch.Tensor()
+
+    labels = np.concatenate(labels, 0)  # labels.shape = (866643, 5) for COCO
+    classes = labels[:, 0].astype(np.int)  # labels = [class xywh]
+    weights = np.bincount(classes, minlength=nc)  # occurences per class
+
+    # Prepend gridpoint count (for uCE trianing)
+    # gpi = ((320 / 32 * np.array([1, 2, 4])) ** 2 * 3).sum()  # gridpoints per image
+    # weights = np.hstack([gpi * len(labels)  - weights.sum() * 9, weights * 9]) ** 0.5  # prepend gridpoints to start
+
+    weights[weights == 0] = 1  # replace empty bins with 1
+    weights = 1 / weights  # number of targets per class
+    weights /= weights.sum()  # normalize
+    return torch.from_numpy(weights)
+
+
+def labels_to_image_weights(labels, nc=80, class_weights=np.ones(80)):
+    # Produces image weights based on class mAPs
+    n = len(labels)
+    class_counts = np.array([np.bincount(labels[i][:, 0].astype(np.int), minlength=nc) for i in range(n)])
+    image_weights = (class_weights.reshape(1, nc) * class_counts).sum(1)
+    # index = random.choices(range(n), weights=image_weights, k=1)  # weight image sample
+    return image_weights
+
+
+def coco80_to_coco91_class():  # converts 80-index (val2014) to 91-index (paper)
+    # https://tech.amikelive.com/node-718/what-object-categories-labels-are-in-coco-dataset/
+    # a = np.loadtxt('data/coco.names', dtype='str', delimiter='\n')
+    # b = np.loadtxt('data/coco_paper.names', dtype='str', delimiter='\n')
+    # x1 = [list(a[i] == b).index(True) + 1 for i in range(80)]  # darknet to coco
+    # x2 = [list(b[i] == a).index(True) if any(b[i] == a) else None for i in range(91)]  # coco to darknet
+    x = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 27, 28, 31, 32, 33, 34,
+         35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63,
+         64, 65, 67, 70, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 84, 85, 86, 87, 88, 89, 90]
+    return x
+
+
+def xyxy2xywh(x):
+    # Convert nx4 boxes from [x1, y1, x2, y2] to [x, y, w, h] where xy1=top-left, xy2=bottom-right
+    y = torch.zeros_like(x) if isinstance(x, torch.Tensor) else np.zeros_like(x)
+    y[:, 0] = (x[:, 0] + x[:, 2]) / 2  # x center
+    y[:, 1] = (x[:, 1] + x[:, 3]) / 2  # y center
+    y[:, 2] = x[:, 2] - x[:, 0]  # width
+    y[:, 3] = x[:, 3] - x[:, 1]  # height
+    return y
+
+
+def xywh2xyxy(x):
+    # Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
+    y = torch.zeros_like(x) if isinstance(x, torch.Tensor) else np.zeros_like(x)
+    y[:, 0] = x[:, 0] - x[:, 2] / 2  # top left x
+    y[:, 1] = x[:, 1] - x[:, 3] / 2  # top left y
+    y[:, 2] = x[:, 0] + x[:, 2] / 2  # bottom right x
+    y[:, 3] = x[:, 1] + x[:, 3] / 2  # bottom right y
+    return y
+
+
+def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
+    # Rescale coords (xyxy) from img1_shape to img0_shape
+    if ratio_pad is None:  # calculate from img0_shape
+        gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1])  # gain  = old / new
+        pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2  # wh padding
+    else:
+        gain = ratio_pad[0][0]
+        pad = ratio_pad[1]
+
+    coords[:, [0, 2]] -= pad[0]  # x padding
+    coords[:, [1, 3]] -= pad[1]  # y padding
+    coords[:, :4] /= gain
+    clip_coords(coords, img0_shape)
+    return coords
+
+
+def clip_coords(boxes, img_shape):
+    # Clip bounding xyxy bounding boxes to image shape (height, width)
+    boxes[:, 0].clamp_(0, img_shape[1])  # x1
+    boxes[:, 1].clamp_(0, img_shape[0])  # y1
+    boxes[:, 2].clamp_(0, img_shape[1])  # x2
+    boxes[:, 3].clamp_(0, img_shape[0])  # y2
+
+
+def ap_per_class(tp, conf, pred_cls, target_cls):
+    """ Compute the average precision, given the recall and precision curves.
+    Source: https://github.com/rafaelpadilla/Object-Detection-Metrics.
+    # Arguments
+        tp:    True positives (nparray, nx1 or nx10).
+        conf:  Objectness value from 0-1 (nparray).
+        pred_cls: Predicted object classes (nparray).
+        target_cls: True object classes (nparray).
+    # Returns
+        The average precision as computed in py-faster-rcnn.
+    """
+
+    # Sort by objectness
+    i = np.argsort(-conf)
+    tp, conf, pred_cls = tp[i], conf[i], pred_cls[i]
+
+    # Find unique classes
+    unique_classes = np.unique(target_cls)
+
+    # Create Precision-Recall curve and compute AP for each class
+    pr_score = 0.1  # score to evaluate P and R https://github.com/ultralytics/yolov3/issues/898
+    s = [unique_classes.shape[0], tp.shape[1]]  # number class, number iou thresholds (i.e. 10 for mAP0.5...0.95)
+    ap, p, r = np.zeros(s), np.zeros(s), np.zeros(s)
+    for ci, c in enumerate(unique_classes):
+        i = pred_cls == c
+        n_gt = (target_cls == c).sum()  # Number of ground truth objects
+        n_p = i.sum()  # Number of predicted objects
+
+        if n_p == 0 or n_gt == 0:
+            continue
+        else:
+            # Accumulate FPs and TPs
+            fpc = (1 - tp[i]).cumsum(0)
+            tpc = tp[i].cumsum(0)
+
+            # Recall
+            recall = tpc / (n_gt + 1e-16)  # recall curve
+            r[ci] = np.interp(-pr_score, -conf[i], recall[:, 0])  # r at pr_score, negative x, xp because xp decreases
+
+            # Precision
+            precision = tpc / (tpc + fpc)  # precision curve
+            p[ci] = np.interp(-pr_score, -conf[i], precision[:, 0])  # p at pr_score
+
+            # AP from recall-precision curve
+            for j in range(tp.shape[1]):
+                ap[ci, j] = compute_ap(recall[:, j], precision[:, j])
+
+            # Plot
+            # fig, ax = plt.subplots(1, 1, figsize=(5, 5))
+            # ax.plot(recall, precision)
+            # ax.set_xlabel('Recall')
+            # ax.set_ylabel('Precision')
+            # ax.set_xlim(0, 1.01)
+            # ax.set_ylim(0, 1.01)
+            # fig.tight_layout()
+            # fig.savefig('PR_curve.png', dpi=300)
+
+    # Compute F1 score (harmonic mean of precision and recall)
+    f1 = 2 * p * r / (p + r + 1e-16)
+
+    return p, r, ap, f1, unique_classes.astype('int32')
+
+
+def compute_ap(recall, precision):
+    """ Compute the average precision, given the recall and precision curves.
+    Source: https://github.com/rbgirshick/py-faster-rcnn.
+    # Arguments
+        recall:    The recall curve (list).
+        precision: The precision curve (list).
+    # Returns
+        The average precision as computed in py-faster-rcnn.
+    """
+
+    # Append sentinel values to beginning and end
+    mrec = np.concatenate(([0.], recall, [min(recall[-1] + 1E-3, 1.)]))
+    mpre = np.concatenate(([0.], precision, [0.]))
+
+    # Compute the precision envelope
+    mpre = np.flip(np.maximum.accumulate(np.flip(mpre)))
+
+    # Integrate area under curve
+    method = 'interp'  # methods: 'continuous', 'interp'
+    if method == 'interp':
+        x = np.linspace(0, 1, 101)  # 101-point interp (COCO)
+        ap = np.trapz(np.interp(x, mrec, mpre), x)  # integrate
+    else:  # 'continuous'
+        i = np.where(mrec[1:] != mrec[:-1])[0]  # points where x axis (recall) changes
+        ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])  # area under curve
+
+    return ap
+
+
+def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False, DIoU=False, CIoU=False):
+    # Returns the IoU of box1 to box2. box1 is 4, box2 is nx4
+    box2 = box2.T
+
+    # Get the coordinates of bounding boxes
+    if x1y1x2y2:  # x1, y1, x2, y2 = box1
+        b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3]
+        b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3]
+    else:  # transform from xywh to xyxy
+        b1_x1, b1_x2 = box1[0] - box1[2] / 2, box1[0] + box1[2] / 2
+        b1_y1, b1_y2 = box1[1] - box1[3] / 2, box1[1] + box1[3] / 2
+        b2_x1, b2_x2 = box2[0] - box2[2] / 2, box2[0] + box2[2] / 2
+        b2_y1, b2_y2 = box2[1] - box2[3] / 2, box2[1] + box2[3] / 2
+
+    # Intersection area
+    inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \
+            (torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0)
+
+    # Union Area
+    w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1
+    w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1
+    union = (w1 * h1 + 1e-16) + w2 * h2 - inter
+
+    iou = inter / union  # iou
+    if GIoU or DIoU or CIoU:
+        cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1)  # convex (smallest enclosing box) width
+        ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1)  # convex height
+        if GIoU:  # Generalized IoU https://arxiv.org/pdf/1902.09630.pdf
+            c_area = cw * ch + 1e-16  # convex area
+            return iou - (c_area - union) / c_area  # GIoU
+        if DIoU or CIoU:  # Distance or Complete IoU https://arxiv.org/abs/1911.08287v1
+            # convex diagonal squared
+            c2 = cw ** 2 + ch ** 2 + 1e-16
+            # centerpoint distance squared
+            rho2 = ((b2_x1 + b2_x2) - (b1_x1 + b1_x2)) ** 2 / 4 + ((b2_y1 + b2_y2) - (b1_y1 + b1_y2)) ** 2 / 4
+            if DIoU:
+                return iou - rho2 / c2  # DIoU
+            elif CIoU:  # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47
+                v = (4 / math.pi ** 2) * torch.pow(torch.atan(w2 / h2) - torch.atan(w1 / h1), 2)
+                with torch.no_grad():
+                    alpha = v / (1 - iou + v + 1e-16)
+                return iou - (rho2 / c2 + v * alpha)  # CIoU
+
+    return iou
+
+
+def box_iou(box1, box2):
+    # https://github.com/pytorch/vision/blob/master/torchvision/ops/boxes.py
+    """
+    Return intersection-over-union (Jaccard index) of boxes.
+    Both sets of boxes are expected to be in (x1, y1, x2, y2) format.
+    Arguments:
+        box1 (Tensor[N, 4])
+        box2 (Tensor[M, 4])
+    Returns:
+        iou (Tensor[N, M]): the NxM matrix containing the pairwise
+            IoU values for every element in boxes1 and boxes2
+    """
+
+    def box_area(box):
+        # box = 4xn
+        return (box[2] - box[0]) * (box[3] - box[1])
+
+    area1 = box_area(box1.T)
+    area2 = box_area(box2.T)
+
+    # inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
+    inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
+    return inter / (area1[:, None] + area2 - inter)  # iou = inter / (area1 + area2 - inter)
+
+
+def wh_iou(wh1, wh2):
+    # Returns the nxm IoU matrix. wh1 is nx2, wh2 is mx2
+    wh1 = wh1[:, None]  # [N,1,2]
+    wh2 = wh2[None]  # [1,M,2]
+    inter = torch.min(wh1, wh2).prod(2)  # [N,M]
+    return inter / (wh1.prod(2) + wh2.prod(2) - inter)  # iou = inter / (area1 + area2 - inter)
+
+
+class FocalLoss(nn.Module):
+    # Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)
+    def __init__(self, loss_fcn, gamma=1.5, alpha=0.25):
+        super(FocalLoss, self).__init__()
+        self.loss_fcn = loss_fcn  # must be nn.BCEWithLogitsLoss()
+        self.gamma = gamma
+        self.alpha = alpha
+        self.reduction = loss_fcn.reduction
+        self.loss_fcn.reduction = 'none'  # required to apply FL to each element
+
+    def forward(self, pred, true):
+        loss = self.loss_fcn(pred, true)
+        # p_t = torch.exp(-loss)
+        # loss *= self.alpha * (1.000001 - p_t) ** self.gamma  # non-zero power for gradient stability
+
+        # TF implementation https://github.com/tensorflow/addons/blob/v0.7.1/tensorflow_addons/losses/focal_loss.py
+        pred_prob = torch.sigmoid(pred)  # prob from logits
+        p_t = true * pred_prob + (1 - true) * (1 - pred_prob)
+        alpha_factor = true * self.alpha + (1 - true) * (1 - self.alpha)
+        modulating_factor = (1.0 - p_t) ** self.gamma
+        loss *= alpha_factor * modulating_factor
+
+        if self.reduction == 'mean':
+            return loss.mean()
+        elif self.reduction == 'sum':
+            return loss.sum()
+        else:  # 'none'
+            return loss
+
+
+def smooth_BCE(eps=0.1):  # https://github.com/ultralytics/yolov3/issues/238#issuecomment-598028441
+    # return positive, negative label smoothing BCE targets
+    return 1.0 - 0.5 * eps, 0.5 * eps
+
+
+class BCEBlurWithLogitsLoss(nn.Module):
+    # BCEwithLogitLoss() with reduced missing label effects.
+    def __init__(self, alpha=0.05):
+        super(BCEBlurWithLogitsLoss, self).__init__()
+        self.loss_fcn = nn.BCEWithLogitsLoss(reduction='none')  # must be nn.BCEWithLogitsLoss()
+        self.alpha = alpha
+
+    def forward(self, pred, true):
+        loss = self.loss_fcn(pred, true)
+        pred = torch.sigmoid(pred)  # prob from logits
+        dx = pred - true  # reduce only missing label effects
+        # dx = (pred - true).abs()  # reduce missing label and false label effects
+        alpha_factor = 1 - torch.exp((dx - 1) / (self.alpha + 1e-4))
+        loss *= alpha_factor
+        return loss.mean()
+
+
+def compute_loss(p, targets, model):  # predictions, targets, model
+    device = targets.device
+    lcls, lbox, lobj = torch.zeros(1, device=device), torch.zeros(1, device=device), torch.zeros(1, device=device)
+    tcls, tbox, indices, anchors = build_targets(p, targets, model)  # targets
+    h = model.hyp  # hyperparameters
+
+    # Define criteria
+    BCEcls = nn.BCEWithLogitsLoss(pos_weight=torch.Tensor([h['cls_pw']])).to(device)
+    BCEobj = nn.BCEWithLogitsLoss(pos_weight=torch.Tensor([h['obj_pw']])).to(device)
+
+    # Class label smoothing https://arxiv.org/pdf/1902.04103.pdf eqn 3
+    cp, cn = smooth_BCE(eps=0.0)
+
+    # Focal loss
+    g = h['fl_gamma']  # focal loss gamma
+    if g > 0:
+        BCEcls, BCEobj = FocalLoss(BCEcls, g), FocalLoss(BCEobj, g)
+
+    # Losses
+    nt = 0  # number of targets
+    np = len(p)  # number of outputs
+    balance = [4.0, 1.0, 0.4] if np == 3 else [4.0, 1.0, 0.4, 0.1]  # P3-5 or P3-6
+    for i, pi in enumerate(p):  # layer index, layer predictions
+        b, a, gj, gi = indices[i]  # image, anchor, gridy, gridx
+        tobj = torch.zeros_like(pi[..., 0], device=device)  # target obj
+
+        n = b.shape[0]  # number of targets
+        if n:
+            nt += n  # cumulative targets
+            ps = pi[b, a, gj, gi]  # prediction subset corresponding to targets
+
+            # Regression
+            pxy = ps[:, :2].sigmoid() * 2. - 0.5
+            pwh = (ps[:, 2:4].sigmoid() * 2) ** 2 * anchors[i]
+            pbox = torch.cat((pxy, pwh), 1).to(device)  # predicted box
+            giou = bbox_iou(pbox.T, tbox[i], x1y1x2y2=False, CIoU=True)  # giou(prediction, target)
+            lbox += (1.0 - giou).mean()  # giou loss
+
+            # Objectness
+            tobj[b, a, gj, gi] = (1.0 - model.gr) + model.gr * giou.detach().clamp(0).type(tobj.dtype)  # giou ratio
+
+            # Classification
+            if model.nc > 1:  # cls loss (only if multiple classes)
+                t = torch.full_like(ps[:, 5:], cn, device=device)  # targets
+                t[range(n), tcls[i]] = cp
+                lcls += BCEcls(ps[:, 5:], t)  # BCE
+
+            # Append targets to text file
+            # with open('targets.txt', 'a') as file:
+            #     [file.write('%11.5g ' * 4 % tuple(x) + '\n') for x in torch.cat((txy[i], twh[i]), 1)]
+
+        lobj += BCEobj(pi[..., 4], tobj) * balance[i]  # obj loss
+
+    s = 3 / np  # output count scaling
+    lbox *= h['giou'] * s
+    lobj *= h['obj'] * s * (1.4 if np == 4 else 1.)
+    lcls *= h['cls'] * s
+    bs = tobj.shape[0]  # batch size
+
+    loss = lbox + lobj + lcls
+    return loss * bs, torch.cat((lbox, lobj, lcls, loss)).detach()
+
+
+def build_targets(p, targets, model):
+    # Build targets for compute_loss(), input targets(image,class,x,y,w,h)
+    det = model.module.model[-1] if is_parallel(model) else model.model[-1]  # Detect() module
+    na, nt = det.na, targets.shape[0]  # number of anchors, targets
+    tcls, tbox, indices, anch = [], [], [], []
+    gain = torch.ones(7, device=targets.device)  # normalized to gridspace gain
+    ai = torch.arange(na, device=targets.device).float().view(na, 1).repeat(1, nt)  # same as .repeat_interleave(nt)
+    targets = torch.cat((targets.repeat(na, 1, 1), ai[:, :, None]), 2)  # append anchor indices
+
+    g = 0.5  # bias
+    off = torch.tensor([[0, 0],
+                        [1, 0], [0, 1], [-1, 0], [0, -1],  # j,k,l,m
+                        # [1, 1], [1, -1], [-1, 1], [-1, -1],  # jk,jm,lk,lm
+                        ], device=targets.device).float() * g  # offsets
+
+    for i in range(det.nl):
+        anchors = det.anchors[i]
+        gain[2:6] = torch.tensor(p[i].shape)[[3, 2, 3, 2]]  # xyxy gain
+
+        # Match targets to anchors
+        t = targets * gain
+        if nt:
+            # Matches
+            r = t[:, :, 4:6] / anchors[:, None]  # wh ratio
+            j = torch.max(r, 1. / r).max(2)[0] < model.hyp['anchor_t']  # compare
+            # j = wh_iou(anchors, t[:, 4:6]) > model.hyp['iou_t']  # iou(3,n)=wh_iou(anchors(3,2), gwh(n,2))
+            t = t[j]  # filter
+
+            # Offsets
+            gxy = t[:, 2:4]  # grid xy
+            gxi = gain[[2, 3]] - gxy  # inverse
+            j, k = ((gxy % 1. < g) & (gxy > 1.)).T
+            l, m = ((gxi % 1. < g) & (gxi > 1.)).T
+            j = torch.stack((torch.ones_like(j), j, k, l, m))
+            t = t.repeat((5, 1, 1))[j]
+            offsets = (torch.zeros_like(gxy)[None] + off[:, None])[j]
+        else:
+            t = targets[0]
+            offsets = 0
+
+        # Define
+        b, c = t[:, :2].long().T  # image, class
+        gxy = t[:, 2:4]  # grid xy
+        gwh = t[:, 4:6]  # grid wh
+        gij = (gxy - offsets).long()
+        gi, gj = gij.T  # grid xy indices
+
+        # Append
+        a = t[:, 6].long()  # anchor indices
+        indices.append((b, a, gj, gi))  # image, anchor, grid indices
+        tbox.append(torch.cat((gxy - gij, gwh), 1))  # box
+        anch.append(anchors[a])  # anchors
+        tcls.append(c)  # class
+
+    return tcls, tbox, indices, anch
+
+
+def non_max_suppression(prediction, conf_thres=0.1, iou_thres=0.6, merge=False, classes=None, agnostic=False):
+    """Performs Non-Maximum Suppression (NMS) on inference results
+
+    Returns:
+         detections with shape: nx6 (x1, y1, x2, y2, conf, cls)
+    """
+    if prediction.dtype is torch.float16:
+        prediction = prediction.float()  # to FP32
+
+    nc = prediction[0].shape[1] - 5  # number of classes
+    xc = prediction[..., 4] > conf_thres  # candidates
+
+    # Settings
+    min_wh, max_wh = 2, 4096  # (pixels) minimum and maximum box width and height
+    max_det = 300  # maximum number of detections per image
+    time_limit = 10.0  # seconds to quit after
+    redundant = True  # require redundant detections
+    multi_label = nc > 1  # multiple labels per box (adds 0.5ms/img)
+
+    t = time.time()
+    output = [None] * prediction.shape[0]
+    for xi, x in enumerate(prediction):  # image index, image inference
+        # Apply constraints
+        # x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0  # width-height
+        x = x[xc[xi]]  # confidence
+
+        # If none remain process next image
+        if not x.shape[0]:
+            continue
+
+        # Compute conf
+        x[:, 5:] *= x[:, 4:5]  # conf = obj_conf * cls_conf
+
+        # Box (center x, center y, width, height) to (x1, y1, x2, y2)
+        box = xywh2xyxy(x[:, :4])
+
+        # Detections matrix nx6 (xyxy, conf, cls)
+        if multi_label:
+            i, j = (x[:, 5:] > conf_thres).nonzero(as_tuple=False).T
+            x = torch.cat((box[i], x[i, j + 5, None], j[:, None].float()), 1)
+        else:  # best class only
+            conf, j = x[:, 5:].max(1, keepdim=True)
+            x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]
+
+        # Filter by class
+        if classes:
+            x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]
+
+        # Apply finite constraint
+        # if not torch.isfinite(x).all():
+        #     x = x[torch.isfinite(x).all(1)]
+
+        # If none remain process next image
+        n = x.shape[0]  # number of boxes
+        if not n:
+            continue
+
+        # Sort by confidence
+        # x = x[x[:, 4].argsort(descending=True)]
+
+        # Batched NMS
+        c = x[:, 5:6] * (0 if agnostic else max_wh)  # classes
+        boxes, scores = x[:, :4] + c, x[:, 4]  # boxes (offset by class), scores
+        i = torchvision.ops.boxes.nms(boxes, scores, iou_thres)
+        if i.shape[0] > max_det:  # limit detections
+            i = i[:max_det]
+        if merge and (1 < n < 3E3):  # Merge NMS (boxes merged using weighted mean)
+            try:  # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
+                iou = box_iou(boxes[i], boxes) > iou_thres  # iou matrix
+                weights = iou * scores[None]  # box weights
+                x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True)  # merged boxes
+                if redundant:
+                    i = i[iou.sum(1) > 1]  # require redundancy
+            except:  # possible CUDA error https://github.com/ultralytics/yolov3/issues/1139
+                print(x, i, x.shape, i.shape)
+                pass
+
+        output[xi] = x[i]
+        if (time.time() - t) > time_limit:
+            break  # time limit exceeded
+
+    return output
+
+
+def strip_optimizer(f='weights/best.pt', s=''):  # from utils.utils import *; strip_optimizer()
+    # Strip optimizer from 'f' to finalize training, optionally save as 's'
+    x = torch.load(f, map_location=torch.device('cpu'))
+    x['optimizer'] = None
+    x['training_results'] = None
+    x['epoch'] = -1
+    x['model'].half()  # to FP16
+    for p in x['model'].parameters():
+        p.requires_grad = False
+    torch.save(x, s or f)
+    mb = os.path.getsize(s or f) / 1E6  # filesize
+    print('Optimizer stripped from %s,%s %.1fMB' % (f, (' saved as %s,' % s) if s else '', mb))
+
+
+def coco_class_count(path='../coco/labels/train2014/'):
+    # Histogram of occurrences per class
+    nc = 80  # number classes
+    x = np.zeros(nc, dtype='int32')
+    files = sorted(glob.glob('%s/*.*' % path))
+    for i, file in enumerate(files):
+        labels = np.loadtxt(file, dtype=np.float32).reshape(-1, 5)
+        x += np.bincount(labels[:, 0].astype('int32'), minlength=nc)
+        print(i, len(files))
+
+
+def coco_only_people(path='../coco/labels/train2017/'):  # from utils.utils import *; coco_only_people()
+    # Find images with only people
+    files = sorted(glob.glob('%s/*.*' % path))
+    for i, file in enumerate(files):
+        labels = np.loadtxt(file, dtype=np.float32).reshape(-1, 5)
+        if all(labels[:, 0] == 0):
+            print(labels.shape[0], file)
+
+
+def crop_images_random(path='../images/', scale=0.50):  # from utils.utils import *; crop_images_random()
+    # crops images into random squares up to scale fraction
+    # WARNING: overwrites images!
+    for file in tqdm(sorted(glob.glob('%s/*.*' % path))):
+        img = cv2.imread(file)  # BGR
+        if img is not None:
+            h, w = img.shape[:2]
+
+            # create random mask
+            a = 30  # minimum size (pixels)
+            mask_h = random.randint(a, int(max(a, h * scale)))  # mask height
+            mask_w = mask_h  # mask width
+
+            # box
+            xmin = max(0, random.randint(0, w) - mask_w // 2)
+            ymin = max(0, random.randint(0, h) - mask_h // 2)
+            xmax = min(w, xmin + mask_w)
+            ymax = min(h, ymin + mask_h)
+
+            # apply random color mask
+            cv2.imwrite(file, img[ymin:ymax, xmin:xmax])
+
+
+def coco_single_class_labels(path='../coco/labels/train2014/', label_class=43):
+    # Makes single-class coco datasets. from utils.utils import *; coco_single_class_labels()
+    if os.path.exists('new/'):
+        shutil.rmtree('new/')  # delete output folder
+    os.makedirs('new/')  # make new output folder
+    os.makedirs('new/labels/')
+    os.makedirs('new/images/')
+    for file in tqdm(sorted(glob.glob('%s/*.*' % path))):
+        with open(file, 'r') as f:
+            labels = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32)
+        i = labels[:, 0] == label_class
+        if any(i):
+            img_file = file.replace('labels', 'images').replace('txt', 'jpg')
+            labels[:, 0] = 0  # reset class to 0
+            with open('new/images.txt', 'a') as f:  # add image to dataset list
+                f.write(img_file + '\n')
+            with open('new/labels/' + Path(file).name, 'a') as f:  # write label
+                for l in labels[i]:
+                    f.write('%g %.6f %.6f %.6f %.6f\n' % tuple(l))
+            shutil.copyfile(src=img_file, dst='new/images/' + Path(file).name.replace('txt', 'jpg'))  # copy images
+
+
+def kmean_anchors(path='./data/coco128.yaml', n=9, img_size=640, thr=4.0, gen=1000, verbose=True):
+    """ Creates kmeans-evolved anchors from training dataset
+
+        Arguments:
+            path: path to dataset *.yaml, or a loaded dataset
+            n: number of anchors
+            img_size: image size used for training
+            thr: anchor-label wh ratio threshold hyperparameter hyp['anchor_t'] used for training, default=4.0
+            gen: generations to evolve anchors using genetic algorithm
+
+        Return:
+            k: kmeans evolved anchors
+
+        Usage:
+            from utils.utils import *; _ = kmean_anchors()
+    """
+    thr = 1. / thr
+
+    def metric(k, wh):  # compute metrics
+        r = wh[:, None] / k[None]
+        x = torch.min(r, 1. / r).min(2)[0]  # ratio metric
+        # x = wh_iou(wh, torch.tensor(k))  # iou metric
+        return x, x.max(1)[0]  # x, best_x
+
+    def fitness(k):  # mutation fitness
+        _, best = metric(torch.tensor(k, dtype=torch.float32), wh)
+        return (best * (best > thr).float()).mean()  # fitness
+
+    def print_results(k):
+        k = k[np.argsort(k.prod(1))]  # sort small to large
+        x, best = metric(k, wh0)
+        bpr, aat = (best > thr).float().mean(), (x > thr).float().mean() * n  # best possible recall, anch > thr
+        print('thr=%.2f: %.4f best possible recall, %.2f anchors past thr' % (thr, bpr, aat))
+        print('n=%g, img_size=%s, metric_all=%.3f/%.3f-mean/best, past_thr=%.3f-mean: ' %
+              (n, img_size, x.mean(), best.mean(), x[x > thr].mean()), end='')
+        for i, x in enumerate(k):
+            print('%i,%i' % (round(x[0]), round(x[1])), end=',  ' if i < len(k) - 1 else '\n')  # use in *.cfg
+        return k
+
+    if isinstance(path, str):  # *.yaml file
+        with open(path) as f:
+            data_dict = yaml.load(f, Loader=yaml.FullLoader)  # model dict
+        from utils.datasets import LoadImagesAndLabels
+        dataset = LoadImagesAndLabels(data_dict['train'], augment=True, rect=True)
+    else:
+        dataset = path  # dataset
+
+    # Get label wh
+    shapes = img_size * dataset.shapes / dataset.shapes.max(1, keepdims=True)
+    wh0 = np.concatenate([l[:, 3:5] * s for s, l in zip(shapes, dataset.labels)])  # wh
+
+    # Filter
+    i = (wh0 < 3.0).any(1).sum()
+    if i:
+        print('WARNING: Extremely small objects found. '
+              '%g of %g labels are < 3 pixels in width or height.' % (i, len(wh0)))
+    wh = wh0[(wh0 >= 2.0).any(1)]  # filter > 2 pixels
+
+    # Kmeans calculation
+    print('Running kmeans for %g anchors on %g points...' % (n, len(wh)))
+    s = wh.std(0)  # sigmas for whitening
+    k, dist = kmeans(wh / s, n, iter=30)  # points, mean distance
+    k *= s
+    wh = torch.tensor(wh, dtype=torch.float32)  # filtered
+    wh0 = torch.tensor(wh0, dtype=torch.float32)  # unflitered
+    k = print_results(k)
+
+    # Plot
+    # k, d = [None] * 20, [None] * 20
+    # for i in tqdm(range(1, 21)):
+    #     k[i-1], d[i-1] = kmeans(wh / s, i)  # points, mean distance
+    # fig, ax = plt.subplots(1, 2, figsize=(14, 7))
+    # ax = ax.ravel()
+    # ax[0].plot(np.arange(1, 21), np.array(d) ** 2, marker='.')
+    # fig, ax = plt.subplots(1, 2, figsize=(14, 7))  # plot wh
+    # ax[0].hist(wh[wh[:, 0]<100, 0],400)
+    # ax[1].hist(wh[wh[:, 1]<100, 1],400)
+    # fig.tight_layout()
+    # fig.savefig('wh.png', dpi=200)
+
+    # Evolve
+    npr = np.random
+    f, sh, mp, s = fitness(k), k.shape, 0.9, 0.1  # fitness, generations, mutation prob, sigma
+    pbar = tqdm(range(gen), desc='Evolving anchors with Genetic Algorithm')  # progress bar
+    for _ in pbar:
+        v = np.ones(sh)
+        while (v == 1).all():  # mutate until a change occurs (prevent duplicates)
+            v = ((npr.random(sh) < mp) * npr.random() * npr.randn(*sh) * s + 1).clip(0.3, 3.0)
+        kg = (k.copy() * v).clip(min=2.0)
+        fg = fitness(kg)
+        if fg > f:
+            f, k = fg, kg.copy()
+            pbar.desc = 'Evolving anchors with Genetic Algorithm: fitness = %.4f' % f
+            if verbose:
+                print_results(k)
+
+    return print_results(k)
+
+
+def print_mutation(hyp, results, yaml_file='hyp_evolved.yaml', bucket=''):
+    # Print mutation results to evolve.txt (for use with train.py --evolve)
+    a = '%10s' * len(hyp) % tuple(hyp.keys())  # hyperparam keys
+    b = '%10.3g' * len(hyp) % tuple(hyp.values())  # hyperparam values
+    c = '%10.4g' * len(results) % results  # results (P, R, [email protected], [email protected]:0.95, val_losses x 3)
+    print('\n%s\n%s\nEvolved fitness: %s\n' % (a, b, c))
+
+    if bucket:
+        os.system('gsutil cp gs://%s/evolve.txt .' % bucket)  # download evolve.txt
+
+    with open('evolve.txt', 'a') as f:  # append result
+        f.write(c + b + '\n')
+    x = np.unique(np.loadtxt('evolve.txt', ndmin=2), axis=0)  # load unique rows
+    x = x[np.argsort(-fitness(x))]  # sort
+    np.savetxt('evolve.txt', x, '%10.3g')  # save sort by fitness
+
+    if bucket:
+        os.system('gsutil cp evolve.txt gs://%s' % bucket)  # upload evolve.txt
+
+    # Save yaml
+    for i, k in enumerate(hyp.keys()):
+        hyp[k] = float(x[0, i + 7])
+    with open(yaml_file, 'w') as f:
+        results = tuple(x[0, :7])
+        c = '%10.4g' * len(results) % results  # results (P, R, [email protected], [email protected]:0.95, val_losses x 3)
+        f.write('# Hyperparameter Evolution Results\n# Generations: %g\n# Metrics: ' % len(x) + c + '\n\n')
+        yaml.dump(hyp, f, sort_keys=False)
+
+
+def apply_classifier(x, model, img, im0):
+    # applies a second stage classifier to yolo outputs
+    im0 = [im0] if isinstance(im0, np.ndarray) else im0
+    for i, d in enumerate(x):  # per image
+        if d is not None and len(d):
+            d = d.clone()
+
+            # Reshape and pad cutouts
+            b = xyxy2xywh(d[:, :4])  # boxes
+            b[:, 2:] = b[:, 2:].max(1)[0].unsqueeze(1)  # rectangle to square
+            b[:, 2:] = b[:, 2:] * 1.3 + 30  # pad
+            d[:, :4] = xywh2xyxy(b).long()
+
+            # Rescale boxes from img_size to im0 size
+            scale_coords(img.shape[2:], d[:, :4], im0[i].shape)
+
+            # Classes
+            pred_cls1 = d[:, 5].long()
+            ims = []
+            for j, a in enumerate(d):  # per item
+                cutout = im0[i][int(a[1]):int(a[3]), int(a[0]):int(a[2])]
+                im = cv2.resize(cutout, (224, 224))  # BGR
+                # cv2.imwrite('test%i.jpg' % j, cutout)
+
+                im = im[:, :, ::-1].transpose(2, 0, 1)  # BGR to RGB, to 3x416x416
+                im = np.ascontiguousarray(im, dtype=np.float32)  # uint8 to float32
+                im /= 255.0  # 0 - 255 to 0.0 - 1.0
+                ims.append(im)
+
+            pred_cls2 = model(torch.Tensor(ims).to(d.device)).argmax(1)  # classifier prediction
+            x[i] = x[i][pred_cls1 == pred_cls2]  # retain matching class detections
+
+    return x
+
+
+def fitness(x):
+    # Returns fitness (for use with results.txt or evolve.txt)
+    w = [0.0, 0.0, 0.1, 0.9]  # weights for [P, R, [email protected], [email protected]:0.95]
+    return (x[:, :4] * w).sum(1)
+
+
+def output_to_target(output, width, height):
+    # Convert model output to target format [batch_id, class_id, x, y, w, h, conf]
+    if isinstance(output, torch.Tensor):
+        output = output.cpu().numpy()
+
+    targets = []
+    for i, o in enumerate(output):
+        if o is not None:
+            for pred in o:
+                box = pred[:4]
+                w = (box[2] - box[0]) / width
+                h = (box[3] - box[1]) / height
+                x = box[0] / width + w / 2
+                y = box[1] / height + h / 2
+                conf = pred[4]
+                cls = int(pred[5])
+
+                targets.append([i, cls, x, y, w, h, conf])
+
+    return np.array(targets)
+
+
+def increment_dir(dir, comment=''):
+    # Increments a directory runs/exp1 --> runs/exp2_comment
+    n = 0  # number
+    dir = str(Path(dir))  # os-agnostic
+    d = sorted(glob.glob(dir + '*'))  # directories
+    if len(d):
+        n = max([int(x[len(dir):x.find('_') if '_' in x else None]) for x in d]) + 1  # increment
+    return dir + str(n) + ('_' + comment if comment else '')
+
+
+# Plotting functions ---------------------------------------------------------------------------------------------------
+def hist2d(x, y, n=100):
+    # 2d histogram used in labels.png and evolve.png
+    xedges, yedges = np.linspace(x.min(), x.max(), n), np.linspace(y.min(), y.max(), n)
+    hist, xedges, yedges = np.histogram2d(x, y, (xedges, yedges))
+    xidx = np.clip(np.digitize(x, xedges) - 1, 0, hist.shape[0] - 1)
+    yidx = np.clip(np.digitize(y, yedges) - 1, 0, hist.shape[1] - 1)
+    return np.log(hist[xidx, yidx])
+
+
+def butter_lowpass_filtfilt(data, cutoff=1500, fs=50000, order=5):
+    # https://stackoverflow.com/questions/28536191/how-to-filter-smooth-with-scipy-numpy
+    def butter_lowpass(cutoff, fs, order):
+        nyq = 0.5 * fs
+        normal_cutoff = cutoff / nyq
+        b, a = butter(order, normal_cutoff, btype='low', analog=False)
+        return b, a
+
+    b, a = butter_lowpass(cutoff, fs, order=order)
+    return filtfilt(b, a, data)  # forward-backward filter
+
+
+def plot_one_box(x, img, color=None, label=None, line_thickness=None):
+    # Plots one bounding box on image img
+    tl = line_thickness or round(0.002 * (img.shape[0] + img.shape[1]) / 2) + 1  # line/font thickness
+    color = color or [random.randint(0, 255) for _ in range(3)]
+    c1, c2 = (int(x[0]), int(x[1])), (int(x[2]), int(x[3]))
+    cv2.rectangle(img, c1, c2, color, thickness=tl, lineType=cv2.LINE_AA)
+    if label:
+        tf = max(tl - 1, 1)  # font thickness
+        t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
+        c2 = c1[0] + t_size[0], c1[1] - t_size[1] - 3
+        cv2.rectangle(img, c1, c2, color, -1, cv2.LINE_AA)  # filled
+        cv2.putText(img, label, (c1[0], c1[1] - 2), 0, tl / 3, [225, 255, 255], thickness=tf, lineType=cv2.LINE_AA)
+
+
+def plot_wh_methods():  # from utils.utils import *; plot_wh_methods()
+    # Compares the two methods for width-height anchor multiplication
+    # https://github.com/ultralytics/yolov3/issues/168
+    x = np.arange(-4.0, 4.0, .1)
+    ya = np.exp(x)
+    yb = torch.sigmoid(torch.from_numpy(x)).numpy() * 2
+
+    fig = plt.figure(figsize=(6, 3), dpi=150)
+    plt.plot(x, ya, '.-', label='YOLOv3')
+    plt.plot(x, yb ** 2, '.-', label='YOLOv5 ^2')
+    plt.plot(x, yb ** 1.6, '.-', label='YOLOv5 ^1.6')
+    plt.xlim(left=-4, right=4)
+    plt.ylim(bottom=0, top=6)
+    plt.xlabel('input')
+    plt.ylabel('output')
+    plt.grid()
+    plt.legend()
+    fig.tight_layout()
+    fig.savefig('comparison.png', dpi=200)
+
+
+def plot_images(images, targets, paths=None, fname='images.jpg', names=None, max_size=640, max_subplots=16):
+    tl = 3  # line thickness
+    tf = max(tl - 1, 1)  # font thickness
+    if os.path.isfile(fname):  # do not overwrite
+        return None
+
+    if isinstance(images, torch.Tensor):
+        images = images.cpu().float().numpy()
+
+    if isinstance(targets, torch.Tensor):
+        targets = targets.cpu().numpy()
+
+    # un-normalise
+    if np.max(images[0]) <= 1:
+        images *= 255
+
+    bs, _, h, w = images.shape  # batch size, _, height, width
+    bs = min(bs, max_subplots)  # limit plot images
+    ns = np.ceil(bs ** 0.5)  # number of subplots (square)
+
+    # Check if we should resize
+    scale_factor = max_size / max(h, w)
+    if scale_factor < 1:
+        h = math.ceil(scale_factor * h)
+        w = math.ceil(scale_factor * w)
+
+    # Empty array for output
+    mosaic = np.full((int(ns * h), int(ns * w), 3), 255, dtype=np.uint8)
+
+    # Fix class - colour map
+    prop_cycle = plt.rcParams['axes.prop_cycle']
+    # https://stackoverflow.com/questions/51350872/python-from-color-name-to-rgb
+    hex2rgb = lambda h: tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))
+    color_lut = [hex2rgb(h) for h in prop_cycle.by_key()['color']]
+
+    for i, img in enumerate(images):
+        if i == max_subplots:  # if last batch has fewer images than we expect
+            break
+
+        block_x = int(w * (i // ns))
+        block_y = int(h * (i % ns))
+
+        img = img.transpose(1, 2, 0)
+        if scale_factor < 1:
+            img = cv2.resize(img, (w, h))
+
+        mosaic[block_y:block_y + h, block_x:block_x + w, :] = img
+        if len(targets) > 0:
+            image_targets = targets[targets[:, 0] == i]
+            boxes = xywh2xyxy(image_targets[:, 2:6]).T
+            classes = image_targets[:, 1].astype('int')
+            gt = image_targets.shape[1] == 6  # ground truth if no conf column
+            conf = None if gt else image_targets[:, 6]  # check for confidence presence (gt vs pred)
+
+            boxes[[0, 2]] *= w
+            boxes[[0, 2]] += block_x
+            boxes[[1, 3]] *= h
+            boxes[[1, 3]] += block_y
+            for j, box in enumerate(boxes.T):
+                cls = int(classes[j])
+                color = color_lut[cls % len(color_lut)]
+                cls = names[cls] if names else cls
+                if gt or conf[j] > 0.3:  # 0.3 conf thresh
+                    label = '%s' % cls if gt else '%s %.1f' % (cls, conf[j])
+                    plot_one_box(box, mosaic, label=label, color=color, line_thickness=tl)
+
+        # Draw image filename labels
+        if paths is not None:
+            label = os.path.basename(paths[i])[:40]  # trim to 40 char
+            t_size = cv2.getTextSize(label, 0, fontScale=tl / 3, thickness=tf)[0]
+            cv2.putText(mosaic, label, (block_x + 5, block_y + t_size[1] + 5), 0, tl / 3, [220, 220, 220], thickness=tf,
+                        lineType=cv2.LINE_AA)
+
+        # Image border
+        cv2.rectangle(mosaic, (block_x, block_y), (block_x + w, block_y + h), (255, 255, 255), thickness=3)
+
+    if fname is not None:
+        mosaic = cv2.resize(mosaic, (int(ns * w * 0.5), int(ns * h * 0.5)), interpolation=cv2.INTER_AREA)
+        cv2.imwrite(fname, cv2.cvtColor(mosaic, cv2.COLOR_BGR2RGB))
+
+    return mosaic
+
+
+def plot_lr_scheduler(optimizer, scheduler, epochs=300, save_dir=''):
+    # Plot LR simulating training for full epochs
+    optimizer, scheduler = copy(optimizer), copy(scheduler)  # do not modify originals
+    y = []
+    for _ in range(epochs):
+        scheduler.step()
+        y.append(optimizer.param_groups[0]['lr'])
+    plt.plot(y, '.-', label='LR')
+    plt.xlabel('epoch')
+    plt.ylabel('LR')
+    plt.grid()
+    plt.xlim(0, epochs)
+    plt.ylim(0)
+    plt.tight_layout()
+    plt.savefig(Path(save_dir) / 'LR.png', dpi=200)
+
+
+def plot_test_txt():  # from utils.utils import *; plot_test()
+    # Plot test.txt histograms
+    x = np.loadtxt('test.txt', dtype=np.float32)
+    box = xyxy2xywh(x[:, :4])
+    cx, cy = box[:, 0], box[:, 1]
+
+    fig, ax = plt.subplots(1, 1, figsize=(6, 6), tight_layout=True)
+    ax.hist2d(cx, cy, bins=600, cmax=10, cmin=0)
+    ax.set_aspect('equal')
+    plt.savefig('hist2d.png', dpi=300)
+
+    fig, ax = plt.subplots(1, 2, figsize=(12, 6), tight_layout=True)
+    ax[0].hist(cx, bins=600)
+    ax[1].hist(cy, bins=600)
+    plt.savefig('hist1d.png', dpi=200)
+
+
+def plot_targets_txt():  # from utils.utils import *; plot_targets_txt()
+    # Plot targets.txt histograms
+    x = np.loadtxt('targets.txt', dtype=np.float32).T
+    s = ['x targets', 'y targets', 'width targets', 'height targets']
+    fig, ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)
+    ax = ax.ravel()
+    for i in range(4):
+        ax[i].hist(x[i], bins=100, label='%.3g +/- %.3g' % (x[i].mean(), x[i].std()))
+        ax[i].legend()
+        ax[i].set_title(s[i])
+    plt.savefig('targets.jpg', dpi=200)
+
+
+def plot_study_txt(f='study.txt', x=None):  # from utils.utils import *; plot_study_txt()
+    # Plot study.txt generated by test.py
+    fig, ax = plt.subplots(2, 4, figsize=(10, 6), tight_layout=True)
+    ax = ax.ravel()
+
+    fig2, ax2 = plt.subplots(1, 1, figsize=(8, 4), tight_layout=True)
+    for f in ['coco_study/study_coco_yolov5%s.txt' % x for x in ['s', 'm', 'l', 'x']]:
+        y = np.loadtxt(f, dtype=np.float32, usecols=[0, 1, 2, 3, 7, 8, 9], ndmin=2).T
+        x = np.arange(y.shape[1]) if x is None else np.array(x)
+        s = ['P', 'R', '[email protected]', '[email protected]:.95', 't_inference (ms/img)', 't_NMS (ms/img)', 't_total (ms/img)']
+        for i in range(7):
+            ax[i].plot(x, y[i], '.-', linewidth=2, markersize=8)
+            ax[i].set_title(s[i])
+
+        j = y[3].argmax() + 1
+        ax2.plot(y[6, :j], y[3, :j] * 1E2, '.-', linewidth=2, markersize=8,
+                 label=Path(f).stem.replace('study_coco_', '').replace('yolo', 'YOLO'))
+
+    ax2.plot(1E3 / np.array([209, 140, 97, 58, 35, 18]), [33.8, 39.6, 43.0, 47.5, 49.4, 50.7],
+             'k.-', linewidth=2, markersize=8, alpha=.25, label='EfficientDet')
+
+    ax2.grid()
+    ax2.set_xlim(0, 30)
+    ax2.set_ylim(28, 50)
+    ax2.set_yticks(np.arange(30, 55, 5))
+    ax2.set_xlabel('GPU Speed (ms/img)')
+    ax2.set_ylabel('COCO AP val')
+    ax2.legend(loc='lower right')
+    plt.savefig('study_mAP_latency.png', dpi=300)
+    plt.savefig(f.replace('.txt', '.png'), dpi=200)
+
+
+def plot_labels(labels, save_dir=''):
+    # plot dataset labels
+    c, b = labels[:, 0], labels[:, 1:].transpose()  # classes, boxes
+    nc = int(c.max() + 1)  # number of classes
+
+    fig, ax = plt.subplots(2, 2, figsize=(8, 8), tight_layout=True)
+    ax = ax.ravel()
+    ax[0].hist(c, bins=np.linspace(0, nc, nc + 1) - 0.5, rwidth=0.8)
+    ax[0].set_xlabel('classes')
+    ax[1].scatter(b[0], b[1], c=hist2d(b[0], b[1], 90), cmap='jet')
+    ax[1].set_xlabel('x')
+    ax[1].set_ylabel('y')
+    ax[2].scatter(b[2], b[3], c=hist2d(b[2], b[3], 90), cmap='jet')
+    ax[2].set_xlabel('width')
+    ax[2].set_ylabel('height')
+    plt.savefig(Path(save_dir) / 'labels.png', dpi=200)
+    plt.close()
+
+
+def plot_evolution(yaml_file='runs/evolve/hyp_evolved.yaml'):  # from utils.utils import *; plot_evolution()
+    # Plot hyperparameter evolution results in evolve.txt
+    with open(yaml_file) as f:
+        hyp = yaml.load(f, Loader=yaml.FullLoader)
+    x = np.loadtxt('evolve.txt', ndmin=2)
+    f = fitness(x)
+    # weights = (f - f.min()) ** 2  # for weighted results
+    plt.figure(figsize=(10, 10), tight_layout=True)
+    matplotlib.rc('font', **{'size': 8})
+    for i, (k, v) in enumerate(hyp.items()):
+        y = x[:, i + 7]
+        # mu = (y * weights).sum() / weights.sum()  # best weighted result
+        mu = y[f.argmax()]  # best single result
+        plt.subplot(5, 5, i + 1)
+        plt.scatter(y, f, c=hist2d(y, f, 20), cmap='viridis', alpha=.8, edgecolors='none')
+        plt.plot(mu, f.max(), 'k+', markersize=15)
+        plt.title('%s = %.3g' % (k, mu), fontdict={'size': 9})  # limit to 40 characters
+        if i % 5 != 0:
+            plt.yticks([])
+        print('%15s: %.3g' % (k, mu))
+    plt.savefig('evolve.png', dpi=200)
+    print('\nPlot saved as evolve.png')
+
+
+def plot_results_overlay(start=0, stop=0):  # from utils.utils import *; plot_results_overlay()
+    # Plot training 'results*.txt', overlaying train and val losses
+    s = ['train', 'train', 'train', 'Precision', '[email protected]', 'val', 'val', 'val', 'Recall', '[email protected]:0.95']  # legends
+    t = ['GIoU', 'Objectness', 'Classification', 'P-R', 'mAP-F1']  # titles
+    for f in sorted(glob.glob('results*.txt') + glob.glob('../../Downloads/results*.txt')):
+        results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
+        n = results.shape[1]  # number of rows
+        x = range(start, min(stop, n) if stop else n)
+        fig, ax = plt.subplots(1, 5, figsize=(14, 3.5), tight_layout=True)
+        ax = ax.ravel()
+        for i in range(5):
+            for j in [i, i + 5]:
+                y = results[j, x]
+                ax[i].plot(x, y, marker='.', label=s[j])
+                # y_smooth = butter_lowpass_filtfilt(y)
+                # ax[i].plot(x, np.gradient(y_smooth), marker='.', label=s[j])
+
+            ax[i].set_title(t[i])
+            ax[i].legend()
+            ax[i].set_ylabel(f) if i == 0 else None  # add filename
+        fig.savefig(f.replace('.txt', '.png'), dpi=200)
+
+
+def plot_results(start=0, stop=0, bucket='', id=(), labels=(),
+                 save_dir=''):  # from utils.utils import *; plot_results()
+    # Plot training 'results*.txt' as seen in https://github.com/ultralytics/yolov5#reproduce-our-training
+    fig, ax = plt.subplots(2, 5, figsize=(12, 6))
+    ax = ax.ravel()
+    s = ['GIoU', 'Objectness', 'Classification', 'Precision', 'Recall',
+         'val GIoU', 'val Objectness', 'val Classification', '[email protected]', '[email protected]:0.95']
+    if bucket:
+        os.system('rm -rf storage.googleapis.com')
+        files = ['https://storage.googleapis.com/%s/results%g.txt' % (bucket, x) for x in id]
+    else:
+        files = glob.glob(str(Path(save_dir) / 'results*.txt')) + glob.glob('../../Downloads/results*.txt')
+    for fi, f in enumerate(files):
+        try:
+            results = np.loadtxt(f, usecols=[2, 3, 4, 8, 9, 12, 13, 14, 10, 11], ndmin=2).T
+            n = results.shape[1]  # number of rows
+            x = range(start, min(stop, n) if stop else n)
+            for i in range(10):
+                y = results[i, x]
+                if i in [0, 1, 2, 5, 6, 7]:
+                    y[y == 0] = np.nan  # dont show zero loss values
+                    # y /= y[0]  # normalize
+                label = labels[fi] if len(labels) else Path(f).stem
+                ax[i].plot(x, y, marker='.', label=label, linewidth=2, markersize=8)
+                ax[i].set_title(s[i])
+                # if i in [5, 6, 7]:  # share train and val loss y axes
+                #     ax[i].get_shared_y_axes().join(ax[i], ax[i - 5])
+        except:
+            print('Warning: Plotting error for %s, skipping file' % f)
+
+    fig.tight_layout()
+    ax[1].legend()
+    fig.savefig(Path(save_dir) / 'results.png', dpi=200)

utils/google_utils.py

@@ -0,0 +1,107 @@
+# This file contains google utils: https://cloud.google.com/storage/docs/reference/libraries
+# pip install --upgrade google-cloud-storage
+# from google.cloud import storage
+
+import os
+import platform
+import time
+from pathlib import Path
+
+
+def attempt_download(weights):
+    # Attempt to download pretrained weights if not found locally
+    weights = weights.strip().replace("'", '')
+    msg = weights + ' missing, try downloading from https://drive.google.com/drive/folders/1Drs_Aiu7xx6S-ix95f9kNsA6ueKRpN2J'
+
+    r = 1  # return
+    if len(weights) > 0 and not os.path.isfile(weights):
+        d = {'yolov3-spp.pt': '1mM67oNw4fZoIOL1c8M3hHmj66d8e-ni_',  # yolov3-spp.yaml
+             'yolov5s.pt': '1R5T6rIyy3lLwgFXNms8whc-387H0tMQO',  # yolov5s.yaml
+             'yolov5m.pt': '1vobuEExpWQVpXExsJ2w-Mbf3HJjWkQJr',  # yolov5m.yaml
+             'yolov5l.pt': '1hrlqD1Wdei7UT4OgT785BEk1JwnSvNEV',  # yolov5l.yaml
+             'yolov5x.pt': '1mM8aZJlWTxOg7BZJvNUMrTnA2AbeCVzS',  # yolov5x.yaml
+             }
+
+        file = Path(weights).name
+        if file in d:
+            r = gdrive_download(id=d[file], name=weights)
+
+        if not (r == 0 and os.path.exists(weights) and os.path.getsize(weights) > 1E6):  # weights exist and > 1MB
+            os.remove(weights) if os.path.exists(weights) else None  # remove partial downloads
+            s = 'curl -L -o %s "storage.googleapis.com/ultralytics/yolov5/ckpt/%s"' % (weights, file)
+            r = os.system(s)  # execute, capture return values
+
+            # Error check
+            if not (r == 0 and os.path.exists(weights) and os.path.getsize(weights) > 1E6):  # weights exist and > 1MB
+                os.remove(weights) if os.path.exists(weights) else None  # remove partial downloads
+                raise Exception(msg)
+
+
+def gdrive_download(id='1n_oKgR81BJtqk75b00eAjdv03qVCQn2f', name='coco128.zip'):
+    # Downloads a file from Google Drive, accepting presented query
+    # from utils.google_utils import *; gdrive_download()
+    t = time.time()
+
+    print('Downloading https://drive.google.com/uc?export=download&id=%s as %s... ' % (id, name), end='')
+    os.remove(name) if os.path.exists(name) else None  # remove existing
+    os.remove('cookie') if os.path.exists('cookie') else None
+
+    # Attempt file download
+    out = "NUL" if platform.system() == "Windows" else "/dev/null"
+    os.system('curl -c ./cookie -s -L "drive.google.com/uc?export=download&id=%s" > %s ' % (id, out))
+    if os.path.exists('cookie'):  # large file
+        s = 'curl -Lb ./cookie "drive.google.com/uc?export=download&confirm=%s&id=%s" -o %s' % (get_token(), id, name)
+    else:  # small file
+        s = 'curl -s -L -o %s "drive.google.com/uc?export=download&id=%s"' % (name, id)
+    r = os.system(s)  # execute, capture return values
+    os.remove('cookie') if os.path.exists('cookie') else None
+
+    # Error check
+    if r != 0:
+        os.remove(name) if os.path.exists(name) else None  # remove partial
+        print('Download error ')  # raise Exception('Download error')
+        return r
+
+    # Unzip if archive
+    if name.endswith('.zip'):
+        print('unzipping... ', end='')
+        os.system('unzip -q %s' % name)  # unzip
+        os.remove(name)  # remove zip to free space
+
+    print('Done (%.1fs)' % (time.time() - t))
+    return r
+
+
+def get_token(cookie="./cookie"):
+    with open(cookie) as f:
+        for line in f:
+            if "download" in line:
+                return line.split()[-1]
+    return ""
+
+# def upload_blob(bucket_name, source_file_name, destination_blob_name):
+#     # Uploads a file to a bucket
+#     # https://cloud.google.com/storage/docs/uploading-objects#storage-upload-object-python
+#
+#     storage_client = storage.Client()
+#     bucket = storage_client.get_bucket(bucket_name)
+#     blob = bucket.blob(destination_blob_name)
+#
+#     blob.upload_from_filename(source_file_name)
+#
+#     print('File {} uploaded to {}.'.format(
+#         source_file_name,
+#         destination_blob_name))
+#
+#
+# def download_blob(bucket_name, source_blob_name, destination_file_name):
+#     # Uploads a blob from a bucket
+#     storage_client = storage.Client()
+#     bucket = storage_client.get_bucket(bucket_name)
+#     blob = bucket.blob(source_blob_name)
+#
+#     blob.download_to_filename(destination_file_name)
+#
+#     print('Blob {} downloaded to {}.'.format(
+#         source_blob_name,
+#         destination_file_name))

utils/torch_utils.py

@@ -0,0 +1,226 @@
+import math
+import os
+import time
+from copy import deepcopy
+
+import torch
+import torch.backends.cudnn as cudnn
+import torch.nn as nn
+import torch.nn.functional as F
+import torchvision.models as models
+
+
+def init_seeds(seed=0):
+    torch.manual_seed(seed)
+
+    # Speed-reproducibility tradeoff https://pytorch.org/docs/stable/notes/randomness.html
+    if seed == 0:  # slower, more reproducible
+        cudnn.deterministic = True
+        cudnn.benchmark = False
+    else:  # faster, less reproducible
+        cudnn.deterministic = False
+        cudnn.benchmark = True
+
+
+def select_device(device='', batch_size=None):
+    # device = 'cpu' or '0' or '0,1,2,3'
+    cpu_request = device.lower() == 'cpu'
+    if device and not cpu_request:  # if device requested other than 'cpu'
+        os.environ['CUDA_VISIBLE_DEVICES'] = device  # set environment variable
+        assert torch.cuda.is_available(), 'CUDA unavailable, invalid device %s requested' % device  # check availablity
+
+    cuda = False if cpu_request else torch.cuda.is_available()
+    if cuda:
+        c = 1024 ** 2  # bytes to MB
+        ng = torch.cuda.device_count()
+        if ng > 1 and batch_size:  # check that batch_size is compatible with device_count
+            assert batch_size % ng == 0, 'batch-size %g not multiple of GPU count %g' % (batch_size, ng)
+        x = [torch.cuda.get_device_properties(i) for i in range(ng)]
+        s = 'Using CUDA '
+        for i in range(0, ng):
+            if i == 1:
+                s = ' ' * len(s)
+            print("%sdevice%g _CudaDeviceProperties(name='%s', total_memory=%dMB)" %
+                  (s, i, x[i].name, x[i].total_memory / c))
+    else:
+        print('Using CPU')
+
+    print('')  # skip a line
+    return torch.device('cuda:0' if cuda else 'cpu')
+
+
+def time_synchronized():
+    torch.cuda.synchronize() if torch.cuda.is_available() else None
+    return time.time()
+
+
+def is_parallel(model):
+    return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel)
+
+
+def intersect_dicts(da, db, exclude=()):
+    # Dictionary intersection of matching keys and shapes, omitting 'exclude' keys, using da values
+    return {k: v for k, v in da.items() if k in db and not any(x in k for x in exclude) and v.shape == db[k].shape}
+
+
+def initialize_weights(model):
+    for m in model.modules():
+        t = type(m)
+        if t is nn.Conv2d:
+            pass  # nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+        elif t is nn.BatchNorm2d:
+            m.eps = 1e-3
+            m.momentum = 0.03
+        elif t in [nn.LeakyReLU, nn.ReLU, nn.ReLU6]:
+            m.inplace = True
+
+
+def find_modules(model, mclass=nn.Conv2d):
+    # Finds layer indices matching module class 'mclass'
+    return [i for i, m in enumerate(model.module_list) if isinstance(m, mclass)]
+
+
+def sparsity(model):
+    # Return global model sparsity
+    a, b = 0., 0.
+    for p in model.parameters():
+        a += p.numel()
+        b += (p == 0).sum()
+    return b / a
+
+
+def prune(model, amount=0.3):
+    # Prune model to requested global sparsity
+    import torch.nn.utils.prune as prune
+    print('Pruning model... ', end='')
+    for name, m in model.named_modules():
+        if isinstance(m, nn.Conv2d):
+            prune.l1_unstructured(m, name='weight', amount=amount)  # prune
+            prune.remove(m, 'weight')  # make permanent
+    print(' %.3g global sparsity' % sparsity(model))
+
+
+def fuse_conv_and_bn(conv, bn):
+    # https://tehnokv.com/posts/fusing-batchnorm-and-conv/
+    with torch.no_grad():
+        # init
+        fusedconv = nn.Conv2d(conv.in_channels,
+                              conv.out_channels,
+                              kernel_size=conv.kernel_size,
+                              stride=conv.stride,
+                              padding=conv.padding,
+                              bias=True).to(conv.weight.device)
+
+        # prepare filters
+        w_conv = conv.weight.clone().view(conv.out_channels, -1)
+        w_bn = torch.diag(bn.weight.div(torch.sqrt(bn.eps + bn.running_var)))
+        fusedconv.weight.copy_(torch.mm(w_bn, w_conv).view(fusedconv.weight.size()))
+
+        # prepare spatial bias
+        b_conv = torch.zeros(conv.weight.size(0), device=conv.weight.device) if conv.bias is None else conv.bias
+        b_bn = bn.bias - bn.weight.mul(bn.running_mean).div(torch.sqrt(bn.running_var + bn.eps))
+        fusedconv.bias.copy_(torch.mm(w_bn, b_conv.reshape(-1, 1)).reshape(-1) + b_bn)
+
+        return fusedconv
+
+
+def model_info(model, verbose=False):
+    # Plots a line-by-line description of a PyTorch model
+    n_p = sum(x.numel() for x in model.parameters())  # number parameters
+    n_g = sum(x.numel() for x in model.parameters() if x.requires_grad)  # number gradients
+    if verbose:
+        print('%5s %40s %9s %12s %20s %10s %10s' % ('layer', 'name', 'gradient', 'parameters', 'shape', 'mu', 'sigma'))
+        for i, (name, p) in enumerate(model.named_parameters()):
+            name = name.replace('module_list.', '')
+            print('%5g %40s %9s %12g %20s %10.3g %10.3g' %
+                  (i, name, p.requires_grad, p.numel(), list(p.shape), p.mean(), p.std()))
+
+    try:  # FLOPS
+        from thop import profile
+        flops = profile(deepcopy(model), inputs=(torch.zeros(1, 3, 64, 64),), verbose=False)[0] / 1E9 * 2
+        fs = ', %.1f GFLOPS' % (flops * 100)  # 640x640 FLOPS
+    except:
+        fs = ''
+
+    print('Model Summary: %g layers, %g parameters, %g gradients%s' % (len(list(model.parameters())), n_p, n_g, fs))
+
+
+def load_classifier(name='resnet101', n=2):
+    # Loads a pretrained model reshaped to n-class output
+    model = models.__dict__[name](pretrained=True)
+
+    # Display model properties
+    input_size = [3, 224, 224]
+    input_space = 'RGB'
+    input_range = [0, 1]
+    mean = [0.485, 0.456, 0.406]
+    std = [0.229, 0.224, 0.225]
+    for x in [input_size, input_space, input_range, mean, std]:
+        print(x + ' =', eval(x))
+
+    # Reshape output to n classes
+    filters = model.fc.weight.shape[1]
+    model.fc.bias = nn.Parameter(torch.zeros(n), requires_grad=True)
+    model.fc.weight = nn.Parameter(torch.zeros(n, filters), requires_grad=True)
+    model.fc.out_features = n
+    return model
+
+
+def scale_img(img, ratio=1.0, same_shape=False):  # img(16,3,256,416), r=ratio
+    # scales img(bs,3,y,x) by ratio
+    if ratio == 1.0:
+        return img
+    else:
+        h, w = img.shape[2:]
+        s = (int(h * ratio), int(w * ratio))  # new size
+        img = F.interpolate(img, size=s, mode='bilinear', align_corners=False)  # resize
+        if not same_shape:  # pad/crop img
+            gs = 32  # (pixels) grid size
+            h, w = [math.ceil(x * ratio / gs) * gs for x in (h, w)]
+        return F.pad(img, [0, w - s[1], 0, h - s[0]], value=0.447)  # value = imagenet mean
+
+
+def copy_attr(a, b, include=(), exclude=()):
+    # Copy attributes from b to a, options to only include [...] and to exclude [...]
+    for k, v in b.__dict__.items():
+        if (len(include) and k not in include) or k.startswith('_') or k in exclude:
+            continue
+        else:
+            setattr(a, k, v)
+
+
+class ModelEMA:
+    """ Model Exponential Moving Average from https://github.com/rwightman/pytorch-image-models
+    Keep a moving average of everything in the model state_dict (parameters and buffers).
+    This is intended to allow functionality like
+    https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage
+    A smoothed version of the weights is necessary for some training schemes to perform well.
+    This class is sensitive where it is initialized in the sequence of model init,
+    GPU assignment and distributed training wrappers.
+    """
+
+    def __init__(self, model, decay=0.9999, updates=0):
+        # Create EMA
+        self.ema = deepcopy(model.module if is_parallel(model) else model).eval()  # FP32 EMA
+        # if next(model.parameters()).device.type != 'cpu':
+        #     self.ema.half()  # FP16 EMA
+        self.updates = updates  # number of EMA updates
+        self.decay = lambda x: decay * (1 - math.exp(-x / 2000))  # decay exponential ramp (to help early epochs)
+        for p in self.ema.parameters():
+            p.requires_grad_(False)
+
+    def update(self, model):
+        # Update EMA parameters
+        with torch.no_grad():
+            self.updates += 1
+            d = self.decay(self.updates)
+
+            msd = model.module.state_dict() if is_parallel(model) else model.state_dict()  # model state_dict
+            for k, v in self.ema.state_dict().items():
+                if v.dtype.is_floating_point:
+                    v *= d
+                    v += (1. - d) * msd[k].detach()
+
+    def update_attr(self, model, include=(), exclude=('process_group', 'reducer')):
+        # Update EMA attributes
+        copy_attr(self.ema, model, include, exclude)