Thiago Hersan
commited on
Commit
β’
988ebda
1
Parent(s):
15a1214
adds app and dockerfile files
Browse files- .gitignore +2 -0
- Dockerfile +12 -0
- README.md +1 -1
- app.py +100 -0
- examples/map-000.jpg +0 -0
- examples/map-010.jpg +0 -0
- examples/map-018.jpg +0 -0
- examples/map-114.jpg +0 -0
- requirements.txt +6 -0
.gitignore
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
.DS_Store
|
2 |
+
|
Dockerfile
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
FROM python:3.8.15
|
2 |
+
|
3 |
+
WORKDIR /code
|
4 |
+
|
5 |
+
COPY ./requirements.txt /code/requirements.txt
|
6 |
+
|
7 |
+
RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
|
8 |
+
|
9 |
+
COPY app.py /code/app.py
|
10 |
+
COPY examples /code/examples
|
11 |
+
|
12 |
+
CMD ["python", "app.py"]
|
README.md
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
---
|
2 |
title: Maskformer Satellite Trees Gradio Docker
|
3 |
-
emoji:
|
4 |
colorFrom: pink
|
5 |
colorTo: green
|
6 |
sdk: docker
|
|
|
1 |
---
|
2 |
title: Maskformer Satellite Trees Gradio Docker
|
3 |
+
emoji: π°π³
|
4 |
colorFrom: pink
|
5 |
colorTo: green
|
6 |
sdk: docker
|
app.py
ADDED
@@ -0,0 +1,100 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import glob
|
2 |
+
import gradio as gr
|
3 |
+
import numpy as np
|
4 |
+
from os import environ
|
5 |
+
from PIL import Image
|
6 |
+
from torchvision import transforms as T
|
7 |
+
from transformers import MaskFormerForInstanceSegmentation, MaskFormerImageProcessor
|
8 |
+
|
9 |
+
|
10 |
+
example_images = sorted(glob.glob('examples/map*.jpg'))
|
11 |
+
|
12 |
+
ade_mean=[0.485, 0.456, 0.406]
|
13 |
+
ade_std=[0.229, 0.224, 0.225]
|
14 |
+
|
15 |
+
test_transform = T.Compose([
|
16 |
+
T.ToTensor(),
|
17 |
+
T.Normalize(mean=ade_mean, std=ade_std)
|
18 |
+
])
|
19 |
+
|
20 |
+
palette = [
|
21 |
+
[120, 120, 120], [4, 200, 4], [4, 4, 250], [6, 230, 230],
|
22 |
+
[80, 50, 50], [120, 120, 80], [140, 140, 140], [204, 5, 255]
|
23 |
+
]
|
24 |
+
|
25 |
+
model_id = f"thiagohersan/maskformer-satellite-trees"
|
26 |
+
vegetation_labels = ["vegetation"]
|
27 |
+
|
28 |
+
# preprocessor = MaskFormerImageProcessor.from_pretrained(model_id)
|
29 |
+
preprocessor = MaskFormerImageProcessor(
|
30 |
+
do_resize=False,
|
31 |
+
do_normalize=False,
|
32 |
+
do_rescale=False,
|
33 |
+
ignore_index=255,
|
34 |
+
reduce_labels=False
|
35 |
+
)
|
36 |
+
|
37 |
+
hf_token = environ.get('HFTOKEN')
|
38 |
+
model = MaskFormerForInstanceSegmentation.from_pretrained(model_id, use_auth_token=hf_token)
|
39 |
+
|
40 |
+
|
41 |
+
def visualize_instance_seg_mask(img_in, mask, id2label, included_labels):
|
42 |
+
img_out = np.zeros((mask.shape[0], mask.shape[1], 3))
|
43 |
+
image_total_pixels = mask.shape[0] * mask.shape[1]
|
44 |
+
label_ids = np.unique(mask)
|
45 |
+
|
46 |
+
id2color = {id: palette[id] for id in label_ids}
|
47 |
+
id2count = {id: 0 for id in label_ids}
|
48 |
+
|
49 |
+
for i in range(img_out.shape[0]):
|
50 |
+
for j in range(img_out.shape[1]):
|
51 |
+
img_out[i, j, :] = id2color[mask[i, j]]
|
52 |
+
id2count[mask[i, j]] = id2count[mask[i, j]] + 1
|
53 |
+
|
54 |
+
image_res = (0.5 * img_in + 0.5 * img_out).astype(np.uint8)
|
55 |
+
|
56 |
+
dataframe = [[
|
57 |
+
f"{id2label[id]}",
|
58 |
+
f"{(100 * id2count[id] / image_total_pixels):.2f} %",
|
59 |
+
f"{np.sqrt(id2count[id] / image_total_pixels):.2f} m"
|
60 |
+
] for id in label_ids if id2label[id] in included_labels]
|
61 |
+
|
62 |
+
if len(dataframe) < 1:
|
63 |
+
dataframe = [[
|
64 |
+
f"",
|
65 |
+
f"{(0):.2f} %",
|
66 |
+
f"{(0):.2f} m"
|
67 |
+
]]
|
68 |
+
|
69 |
+
return image_res, dataframe
|
70 |
+
|
71 |
+
|
72 |
+
def query_image(image_path):
|
73 |
+
img = np.array(Image.open(image_path))
|
74 |
+
img_size = (img.shape[0], img.shape[1])
|
75 |
+
inputs = preprocessor(images=test_transform(img), return_tensors="pt")
|
76 |
+
outputs = model(**inputs)
|
77 |
+
results = preprocessor.post_process_semantic_segmentation(outputs=outputs, target_sizes=[img_size])[0]
|
78 |
+
mask_img, dataframe = visualize_instance_seg_mask(img, results.numpy(), model.config.id2label, vegetation_labels)
|
79 |
+
return mask_img, dataframe
|
80 |
+
|
81 |
+
|
82 |
+
demo = gr.Interface(
|
83 |
+
title="Maskformer Satellite+Trees",
|
84 |
+
description="Using a finetuned version of the [facebook/maskformer-swin-base-ade](https://huggingface.co/facebook/maskformer-swin-base-ade) model (created specifically to work with satellite images) to calculate percentage of pixels in an image that belong to vegetation.",
|
85 |
+
|
86 |
+
fn=query_image,
|
87 |
+
inputs=[gr.Image(type="filepath", label="Input Image")],
|
88 |
+
outputs=[
|
89 |
+
gr.Image(label="Vegetation"),
|
90 |
+
gr.DataFrame(label="Info", headers=["Object Label", "Pixel Percent", "Square Length"])
|
91 |
+
],
|
92 |
+
|
93 |
+
examples=example_images,
|
94 |
+
cache_examples=True,
|
95 |
+
|
96 |
+
allow_flagging="never",
|
97 |
+
analytics_enabled=None
|
98 |
+
)
|
99 |
+
|
100 |
+
demo.launch(show_api=False)
|
examples/map-000.jpg
ADDED
examples/map-010.jpg
ADDED
examples/map-018.jpg
ADDED
examples/map-114.jpg
ADDED
requirements.txt
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio==3.16.2
|
2 |
+
Pillow==9.4.0
|
3 |
+
scipy==1.9.3
|
4 |
+
torch==1.13.1
|
5 |
+
torchvision==0.14.1
|
6 |
+
transformers==4.25.1
|