Update app.py
Browse files
app.py
CHANGED
@@ -1,80 +1,82 @@
|
|
1 |
import torch
|
2 |
from torchvision import models, transforms
|
3 |
-
from PIL import Image
|
4 |
import gradio as gr
|
5 |
from typing import Union
|
6 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
class Preprocessor:
|
8 |
-
def __init__(self
|
9 |
-
self.transform = transforms.Compose([
|
10 |
transforms.ToTensor(),
|
11 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
12 |
-
])
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
class SegmentationModel:
|
18 |
-
def __init__(self):
|
19 |
-
self.model = models.segmentation.deeplabv3_resnet101(pretrained=True)
|
20 |
-
self.model.eval()
|
21 |
-
if torch.cuda.is_available():
|
22 |
-
self.model.to('cuda')
|
23 |
-
|
24 |
-
def predict(self, input_batch: torch.Tensor) -> torch.Tensor:
|
25 |
-
with torch.no_grad():
|
26 |
-
if torch.cuda.is_available():
|
27 |
-
input_batch = input_batch.to('cuda')
|
28 |
-
output: torch.Tensor = self.model(input_batch)['out'][0]
|
29 |
-
return output
|
30 |
|
31 |
-
class
|
32 |
def __init__(self):
|
33 |
palette = torch.tensor([2 ** 25 - 1, 2 ** 15 - 1, 2 ** 21 - 1])
|
34 |
colors = torch.as_tensor([i for i in range(21)])[:, None] * palette
|
35 |
self.colors = (colors % 255).numpy().astype("uint8")
|
36 |
-
|
37 |
-
def
|
38 |
-
|
|
|
39 |
colorized_output.putpalette(self.colors.ravel())
|
40 |
return colorized_output
|
41 |
|
42 |
-
class
|
43 |
-
def __init__(self):
|
44 |
-
self.
|
45 |
-
self.
|
46 |
-
self.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
input_batch: torch.Tensor = input_tensor.unsqueeze(0)
|
52 |
-
output: torch.Tensor = self.model.predict(input_batch)
|
53 |
-
output_predictions: torch.Tensor = output.argmax(0)
|
54 |
-
return self.colorizer.colorize(output_predictions)
|
55 |
|
56 |
class GradioApp:
|
57 |
-
def __init__(self,
|
58 |
-
self.
|
59 |
-
|
60 |
def launch(self):
|
61 |
with gr.Blocks() as demo:
|
62 |
gr.Markdown("<h1 style='text-align: center; color: #4CAF50;'>Deeplabv3 Segmentation</h1>")
|
63 |
gr.Markdown("<p style='text-align: center;'>Upload an image to perform semantic segmentation using Deeplabv3 ResNet101.</p>")
|
64 |
-
gr.Markdown("""
|
65 |
-
### Model Information
|
66 |
-
**DeepLabv3 with ResNet101** is a convolutional neural network model designed for semantic image segmentation.
|
67 |
-
It utilizes atrous convolution to capture multi-scale context by using different atrous rates.
|
68 |
-
""")
|
69 |
with gr.Row():
|
70 |
with gr.Column():
|
71 |
-
|
|
|
|
|
72 |
with gr.Column():
|
73 |
-
|
74 |
-
|
75 |
-
button = gr.Button("Segment")
|
76 |
-
button.click(fn=self.segmenter.segment, inputs=image_input, outputs=image_output)
|
77 |
-
|
78 |
gr.Markdown("### Example Images")
|
79 |
gr.Examples(
|
80 |
examples=[
|
@@ -82,14 +84,16 @@ class GradioApp:
|
|
82 |
["https://www.timeforkids.com/wp-content/uploads/2023/09/G3G5_230915_puffins_on_the_rise.jpg?w=1024"],
|
83 |
["https://www.timeforkids.com/wp-content/uploads/2024/03/G3G5_240412_bug_eyed.jpg?w=1024"]
|
84 |
],
|
85 |
-
inputs=
|
86 |
-
outputs=
|
87 |
label="Click an example to use it"
|
88 |
)
|
89 |
-
|
90 |
demo.launch()
|
91 |
|
92 |
if __name__ == "__main__":
|
93 |
-
|
94 |
-
|
|
|
|
|
|
|
95 |
app.launch()
|
|
|
1 |
import torch
|
2 |
from torchvision import models, transforms
|
3 |
+
from PIL import Image, ImageDraw
|
4 |
import gradio as gr
|
5 |
from typing import Union
|
6 |
|
7 |
+
SEGMENTATION_MODELS = {
|
8 |
+
"deeplabv3_resnet101": models.segmentation.deeplabv3_resnet101,}
|
9 |
+
|
10 |
+
class ModelLoader:
|
11 |
+
def __init__(self, model_dict: dict):
|
12 |
+
self.model_dict = model_dict
|
13 |
+
|
14 |
+
def load_model(self, model_name: str) -> torch.nn.Module:
|
15 |
+
model_name_lower = model_name.lower()
|
16 |
+
if model_name_lower in self.model_dict:
|
17 |
+
model_class = self.model_dict[model_name_lower]
|
18 |
+
model = model_class(pretrained=True)
|
19 |
+
model.eval()
|
20 |
+
return model
|
21 |
+
else:
|
22 |
+
raise ValueError(f"Model {model_name} is not supported")
|
23 |
+
|
24 |
class Preprocessor:
|
25 |
+
def __init__(self, transform: transforms.Compose = transforms.Compose([
|
|
|
26 |
transforms.ToTensor(),
|
27 |
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
28 |
+
])):
|
29 |
+
self.transform = transform
|
30 |
+
|
31 |
+
def preprocess(self, image: Image.Image) -> torch.Tensor:
|
32 |
+
return self.transform(image).unsqueeze(0)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
+
class Postprocessor:
|
35 |
def __init__(self):
|
36 |
palette = torch.tensor([2 ** 25 - 1, 2 ** 15 - 1, 2 ** 21 - 1])
|
37 |
colors = torch.as_tensor([i for i in range(21)])[:, None] * palette
|
38 |
self.colors = (colors % 255).numpy().astype("uint8")
|
39 |
+
|
40 |
+
def postprocess(self, output: torch.Tensor) -> Image.Image:
|
41 |
+
output_predictions = output.argmax(0)
|
42 |
+
colorized_output = Image.fromarray(output_predictions.byte().cpu().numpy(), mode='P')
|
43 |
colorized_output.putpalette(self.colors.ravel())
|
44 |
return colorized_output
|
45 |
|
46 |
+
class Segmentation:
|
47 |
+
def __init__(self, model_loader: ModelLoader, preprocessor: Preprocessor, postprocessor: Postprocessor):
|
48 |
+
self.model_loader = model_loader
|
49 |
+
self.preprocessor = preprocessor
|
50 |
+
self.postprocessor = postprocessor
|
51 |
+
|
52 |
+
def segment(self, image: Image.Image, selected_model: str) -> Image.Image:
|
53 |
+
model = self.model_loader.load_model(selected_model)
|
54 |
+
input_tensor = self.preprocessor.preprocess(image)
|
55 |
+
|
56 |
+
if torch.cuda.is_available():
|
57 |
+
input_tensor = input_tensor.to("cuda")
|
58 |
+
model = model.to("cuda")
|
59 |
|
60 |
+
with torch.no_grad():
|
61 |
+
output = model(input_tensor)['out'][0]
|
62 |
+
return self.postprocessor.postprocess(output)
|
|
|
|
|
|
|
|
|
63 |
|
64 |
class GradioApp:
|
65 |
+
def __init__(self, segmentation: Segmentation):
|
66 |
+
self.segmentation = segmentation
|
67 |
+
|
68 |
def launch(self):
|
69 |
with gr.Blocks() as demo:
|
70 |
gr.Markdown("<h1 style='text-align: center; color: #4CAF50;'>Deeplabv3 Segmentation</h1>")
|
71 |
gr.Markdown("<p style='text-align: center;'>Upload an image to perform semantic segmentation using Deeplabv3 ResNet101.</p>")
|
|
|
|
|
|
|
|
|
|
|
72 |
with gr.Row():
|
73 |
with gr.Column():
|
74 |
+
upload_image = gr.Image(type='pil', label="Upload Image")
|
75 |
+
self.model_dropdown = gr.Dropdown(choices=list(SEGMENTATION_MODELS.keys()), label="Select Model")
|
76 |
+
segment_button = gr.Button("Segment")
|
77 |
with gr.Column():
|
78 |
+
output_image = gr.Image(type='pil', label="Segmented Output")
|
79 |
+
segment_button.click(fn=self.segmentation.segment, inputs=[upload_image, self.model_dropdown], outputs=output_image)
|
|
|
|
|
|
|
80 |
gr.Markdown("### Example Images")
|
81 |
gr.Examples(
|
82 |
examples=[
|
|
|
84 |
["https://www.timeforkids.com/wp-content/uploads/2023/09/G3G5_230915_puffins_on_the_rise.jpg?w=1024"],
|
85 |
["https://www.timeforkids.com/wp-content/uploads/2024/03/G3G5_240412_bug_eyed.jpg?w=1024"]
|
86 |
],
|
87 |
+
inputs=upload_image,
|
88 |
+
outputs=output_image,
|
89 |
label="Click an example to use it"
|
90 |
)
|
|
|
91 |
demo.launch()
|
92 |
|
93 |
if __name__ == "__main__":
|
94 |
+
model_loader = ModelLoader(SEGMENTATION_MODELS)
|
95 |
+
preprocessor = Preprocessor()
|
96 |
+
postprocessor = Postprocessor()
|
97 |
+
segmentation = Segmentation(model_loader, preprocessor, postprocessor)
|
98 |
+
app = GradioApp(segmentation)
|
99 |
app.launch()
|