Update
Browse files
app.py
CHANGED
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from PIL import Image
|
3 |
+
import torch
|
4 |
+
import torch.nn.functional as F
|
5 |
+
import numpy as np
|
6 |
+
from torchvision import models, transforms
|
7 |
+
from pytorch_grad_cam import GradCAM
|
8 |
+
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
|
9 |
+
from pytorch_grad_cam.utils.image import show_cam_on_image
|
10 |
+
|
11 |
+
# Set device (CPU for Hugging Face Spaces)
|
12 |
+
device = torch.device("cpu")
|
13 |
+
|
14 |
+
# Load model
|
15 |
+
model = models.resnet50(weights=None)
|
16 |
+
model.fc = torch.nn.Linear(model.fc.in_features, 2)
|
17 |
+
model.load_state_dict(torch.load("resnet50_dr_classifier.pth", map_location=device))
|
18 |
+
model.to(device)
|
19 |
+
model.eval()
|
20 |
+
|
21 |
+
# Grad-CAM setup
|
22 |
+
target_layer = model.layer4[-1]
|
23 |
+
cam = GradCAM(model=model, target_layers=[target_layer])
|
24 |
+
|
25 |
+
# Image transform
|
26 |
+
transform = transforms.Compose([
|
27 |
+
transforms.Resize((224, 224)),
|
28 |
+
transforms.ToTensor(),
|
29 |
+
transforms.Normalize([0.485, 0.456, 0.406],
|
30 |
+
[0.229, 0.224, 0.225])
|
31 |
+
])
|
32 |
+
|
33 |
+
def predict_retinopathy(image):
|
34 |
+
img = image.convert("RGB").resize((224, 224))
|
35 |
+
img_tensor = transform(img).unsqueeze(0).to(device)
|
36 |
+
|
37 |
+
with torch.no_grad():
|
38 |
+
output = model(img_tensor)
|
39 |
+
probs = F.softmax(output, dim=1)
|
40 |
+
pred = torch.argmax(probs, dim=1).item()
|
41 |
+
confidence = probs[0][pred].item()
|
42 |
+
|
43 |
+
label = "Diabetic Retinopathy (DR)" if pred == 0 else "No DR"
|
44 |
+
|
45 |
+
# Grad-CAM
|
46 |
+
rgb_img_np = np.array(img).astype(np.float32) / 255.0
|
47 |
+
rgb_img_np = np.ascontiguousarray(rgb_img_np)
|
48 |
+
grayscale_cam = cam(input_tensor=img_tensor, targets=[ClassifierOutputTarget(pred)])[0]
|
49 |
+
cam_image = show_cam_on_image(rgb_img_np, grayscale_cam, use_rgb=True)
|
50 |
+
|
51 |
+
cam_pil = Image.fromarray(cam_image)
|
52 |
+
return cam_pil, f"{label} (Confidence: {confidence:.2f})"
|
53 |
+
|
54 |
+
# Gradio UI
|
55 |
+
gr.Interface(
|
56 |
+
fn=predict_retinopathy,
|
57 |
+
inputs=gr.Image(type="pil"),
|
58 |
+
outputs=[
|
59 |
+
gr.Image(type="pil", label="Grad-CAM"),
|
60 |
+
gr.Text(label="Prediction")
|
61 |
+
],
|
62 |
+
title="Diabetic Retinopathy Detection",
|
63 |
+
description="Upload a retinal image to classify DR and view Grad-CAM heatmap."
|
64 |
+
).launch()
|