Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -10,10 +10,19 @@ from torchvision import transforms
|
|
10 |
import torchvision.transforms.functional as TF
|
11 |
import urllib.request
|
12 |
import os
|
|
|
|
|
13 |
|
14 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
15 |
model = None
|
16 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
# Define your Attention U-Net architecture (from your training code)
|
18 |
class DoubleConv(nn.Module):
|
19 |
def __init__(self, in_channels, out_channels):
|
@@ -65,12 +74,10 @@ class AttentionUNET(nn.Module):
|
|
65 |
self.downs = nn.ModuleList()
|
66 |
self.attentions = nn.ModuleList()
|
67 |
self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
|
68 |
-
|
69 |
# Down part of UNET
|
70 |
for feature in features:
|
71 |
self.downs.append(DoubleConv(in_channels, feature))
|
72 |
in_channels = feature
|
73 |
-
|
74 |
# Bottleneck
|
75 |
self.bottleneck = DoubleConv(features[-1], features[-1]*2)
|
76 |
|
@@ -79,19 +86,15 @@ class AttentionUNET(nn.Module):
|
|
79 |
self.ups.append(nn.ConvTranspose2d(feature*2, feature, kernel_size=2, stride=2))
|
80 |
self.attentions.append(AttentionBlock(F_g=feature, F_l=feature, F_int=feature // 2))
|
81 |
self.ups.append(DoubleConv(feature*2, feature))
|
82 |
-
|
83 |
self.final_conv = nn.Conv2d(features[0], out_channels, kernel_size=1)
|
84 |
-
|
85 |
def forward(self, x):
|
86 |
skip_connections = []
|
87 |
for down in self.downs:
|
88 |
x = down(x)
|
89 |
skip_connections.append(x)
|
90 |
x = self.pool(x)
|
91 |
-
|
92 |
x = self.bottleneck(x)
|
93 |
skip_connections = skip_connections[::-1] #reverse list
|
94 |
-
|
95 |
for idx in range(0, len(self.ups), 2): #do up and double_conv
|
96 |
x = self.ups[idx](x)
|
97 |
skip_connection = skip_connections[idx//2]
|
@@ -100,7 +103,6 @@ class AttentionUNET(nn.Module):
|
|
100 |
skip_connection = self.attentions[idx // 2](skip_connection, x)
|
101 |
concat_skip = torch.cat((skip_connection, x), dim=1)
|
102 |
x = self.ups[idx+1](concat_skip)
|
103 |
-
|
104 |
return self.final_conv(x)
|
105 |
|
106 |
def download_model():
|
@@ -109,15 +111,15 @@ def download_model():
|
|
109 |
model_path = "best_attention_model.pth.tar"
|
110 |
|
111 |
if not os.path.exists(model_path):
|
112 |
-
print("
|
113 |
try:
|
114 |
urllib.request.urlretrieve(model_url, model_path)
|
115 |
-
print("
|
116 |
except Exception as e:
|
117 |
-
print(f"
|
118 |
return None
|
119 |
else:
|
120 |
-
print("
|
121 |
|
122 |
return model_path
|
123 |
|
@@ -126,7 +128,7 @@ def load_your_attention_model():
|
|
126 |
global model
|
127 |
if model is None:
|
128 |
try:
|
129 |
-
print("
|
130 |
|
131 |
# Download model if needed
|
132 |
model_path = download_model()
|
@@ -141,9 +143,9 @@ def load_your_attention_model():
|
|
141 |
model.load_state_dict(checkpoint["state_dict"])
|
142 |
model.eval()
|
143 |
|
144 |
-
print("
|
145 |
except Exception as e:
|
146 |
-
print(f"
|
147 |
model = None
|
148 |
return model
|
149 |
|
@@ -161,75 +163,63 @@ def preprocess_for_your_model(image):
|
|
161 |
|
162 |
return val_test_transform(image).unsqueeze(0) # Add batch dimension
|
163 |
|
164 |
-
def
|
165 |
-
"""Create heatmap visualization from continuous prediction values"""
|
166 |
-
# pred_mask_continuous should be the raw sigmoid output (0-1 values)
|
167 |
-
heatmap_np = pred_mask_continuous.cpu().squeeze().numpy()
|
168 |
-
|
169 |
-
# Normalize to 0-255 for better visualization
|
170 |
-
heatmap_normalized = (heatmap_np * 255).astype(np.uint8)
|
171 |
-
|
172 |
-
# Apply colormap (using 'hot' colormap like in medical imaging)
|
173 |
-
heatmap_colored = cv2.applyColorMap(heatmap_normalized, cv2.COLORMAP_HOT)
|
174 |
-
heatmap_colored = cv2.cvtColor(heatmap_colored, cv2.COLOR_BGR2RGB)
|
175 |
-
|
176 |
-
# Convert original image to RGB for overlay
|
177 |
-
if len(original_image.shape) == 2: # Grayscale
|
178 |
-
original_rgb = cv2.cvtColor(original_image.astype(np.uint8), cv2.COLOR_GRAY2RGB)
|
179 |
-
else:
|
180 |
-
original_rgb = original_image.astype(np.uint8)
|
181 |
-
|
182 |
-
# Create overlay (blend original image with heatmap)
|
183 |
-
alpha = 0.6 # Transparency factor
|
184 |
-
overlay = cv2.addWeighted(original_rgb, 1-alpha, heatmap_colored, alpha, 0)
|
185 |
-
|
186 |
-
return overlay
|
187 |
-
|
188 |
-
def predict_tumor(image):
|
189 |
current_model = load_your_attention_model()
|
190 |
|
191 |
if current_model is None:
|
192 |
-
return None, "
|
193 |
if image is None:
|
194 |
-
return None, "
|
195 |
|
196 |
try:
|
197 |
-
print("
|
198 |
|
199 |
# Use the exact preprocessing from your Colab code
|
200 |
input_tensor = preprocess_for_your_model(image).to(device)
|
201 |
|
202 |
# Predict using your model (exactly like your Colab code)
|
203 |
with torch.no_grad():
|
204 |
-
|
205 |
-
pred_mask_binary = (
|
206 |
|
207 |
-
# Convert to numpy (like your Colab code)
|
208 |
pred_mask_np = pred_mask_binary.cpu().squeeze().numpy()
|
|
|
209 |
original_np = np.array(image.convert('L').resize((256, 256)))
|
210 |
|
211 |
-
# Create inverted mask for visualization (like your Colab code)
|
212 |
inv_pred_mask_np = np.where(pred_mask_np == 1, 0, 255)
|
213 |
|
214 |
-
# Create tumor-only image (like your Colab code)
|
215 |
tumor_only = np.where(pred_mask_np == 1, original_np, 255)
|
216 |
|
217 |
-
#
|
218 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
219 |
|
220 |
-
# Create visualization
|
221 |
fig, axes = plt.subplots(1, 5, figsize=(25, 5))
|
222 |
-
fig.suptitle('
|
223 |
|
224 |
-
titles = ["Original Image", "
|
225 |
-
images = [original_np,
|
226 |
-
cmaps = ['gray', '
|
227 |
|
228 |
for i, ax in enumerate(axes):
|
229 |
-
|
230 |
-
ax.imshow(images[i], cmap=cmaps[i])
|
231 |
-
else:
|
232 |
-
ax.imshow(images[i]) # RGB image
|
233 |
ax.set_title(titles[i], fontsize=12, fontweight='bold')
|
234 |
ax.axis('off')
|
235 |
|
@@ -243,68 +233,73 @@ def predict_tumor(image):
|
|
243 |
|
244 |
result_image = Image.open(buf)
|
245 |
|
246 |
-
# Calculate statistics (like your Colab code)
|
247 |
tumor_pixels = np.sum(pred_mask_np)
|
248 |
total_pixels = pred_mask_np.size
|
249 |
tumor_percentage = (tumor_pixels / total_pixels) * 100
|
250 |
|
251 |
# Calculate confidence metrics
|
252 |
-
max_confidence = torch.max(
|
253 |
-
mean_confidence = torch.mean(
|
254 |
|
255 |
analysis_text = f"""
|
256 |
-
##
|
257 |
-
###
|
258 |
-
- **Status**: {'
|
259 |
- **Tumor Area**: {tumor_percentage:.2f}% of brain region
|
260 |
- **Tumor Pixels**: {tumor_pixels:,} pixels
|
261 |
- **Max Confidence**: {max_confidence:.4f}
|
262 |
- **Mean Confidence**: {mean_confidence:.4f}
|
263 |
-
|
264 |
-
|
265 |
-
|
266 |
-
- **
|
267 |
-
- **
|
268 |
-
|
269 |
-
|
270 |
-
###
|
271 |
- **Architecture**: YOUR trained Attention U-Net
|
272 |
- **Training Performance**: Dice: 0.8420, IoU: 0.7297
|
273 |
- **Input**: Grayscale (single channel)
|
274 |
-
- **Output**: Binary segmentation mask
|
275 |
- **Device**: {device.type.upper()}
|
276 |
-
|
277 |
-
### 🎯 Model Performance:
|
278 |
- **Training Accuracy**: 98.90%
|
279 |
- **Best Dice Score**: 0.8420
|
280 |
- **Best IoU Score**: 0.7297
|
281 |
- **Training Dataset**: Brain tumor segmentation dataset
|
282 |
-
|
283 |
-
### 📈 Processing Details:
|
284 |
- **Preprocessing**: Resize(256×256) + ToTensor (your exact method)
|
285 |
- **Threshold**: 0.5 (sigmoid > 0.5)
|
286 |
- **Architecture**: Attention gates + Skip connections
|
287 |
- **Features**: [32, 64, 128, 256] channels
|
288 |
-
|
289 |
-
|
290 |
-
### ⚠️ Medical Disclaimer:
|
291 |
This is YOUR trained AI model for **research and educational purposes only**.
|
292 |
Results should be validated by medical professionals. Not for clinical diagnosis.
|
293 |
-
|
294 |
-
|
295 |
-
✅ This is your own trained model with proven {tumor_percentage:.2f}% detection capability!
|
296 |
"""
|
297 |
|
298 |
-
print(f"
|
299 |
return result_image, analysis_text
|
300 |
|
301 |
except Exception as e:
|
302 |
-
error_msg = f"
|
303 |
print(error_msg)
|
304 |
return None, error_msg
|
305 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
306 |
def clear_all():
|
307 |
-
return None, None, "Upload a brain MRI image to test YOUR trained Attention U-Net model
|
308 |
|
309 |
# Enhanced CSS for your model
|
310 |
css = """
|
@@ -324,13 +319,13 @@ css = """
|
|
324 |
"""
|
325 |
|
326 |
# Create Gradio interface for your model
|
327 |
-
with gr.Blocks(css=css, title="
|
328 |
|
329 |
gr.HTML("""
|
330 |
<div id="title">
|
331 |
-
<h1
|
332 |
<p style="font-size: 18px; margin-top: 15px;">
|
333 |
-
Using Your Own Trained Model • Dice: 0.8420 • IoU: 0.7297
|
334 |
</p>
|
335 |
<p style="font-size: 14px; margin-top: 10px; opacity: 0.9;">
|
336 |
Loaded from: ArchCoder/the-op-segmenter HuggingFace Space
|
@@ -338,9 +333,11 @@ with gr.Blocks(css=css, title="🧠 Your Attention U-Net Model with Heatmap", th
|
|
338 |
</div>
|
339 |
""")
|
340 |
|
|
|
|
|
341 |
with gr.Row():
|
342 |
with gr.Column(scale=1):
|
343 |
-
gr.Markdown("###
|
344 |
|
345 |
image_input = gr.Image(
|
346 |
label="Brain MRI Scan",
|
@@ -350,51 +347,49 @@ with gr.Blocks(css=css, title="🧠 Your Attention U-Net Model with Heatmap", th
|
|
350 |
)
|
351 |
|
352 |
with gr.Row():
|
353 |
-
analyze_btn = gr.Button("
|
354 |
-
|
|
|
355 |
|
356 |
gr.HTML("""
|
357 |
<div style="margin-top: 20px; padding: 20px; background: linear-gradient(135deg, #F3E8FF 0%, #EDE9FE 100%); border-radius: 10px; border-left: 4px solid #8B5CF6;">
|
358 |
-
<h4 style="color: #8B5CF6; margin-bottom: 15px;"
|
359 |
<ul style="margin: 10px 0; padding-left: 20px; line-height: 1.6;">
|
360 |
<li><strong>Personal Model:</strong> Your own trained Attention U-Net</li>
|
361 |
<li><strong>Proven Performance:</strong> 84.2% Dice Score, 72.97% IoU</li>
|
362 |
<li><strong>Attention Gates:</strong> Advanced feature selection</li>
|
363 |
<li><strong>Clean Output:</strong> Binary segmentation masks</li>
|
364 |
-
<li><strong>
|
365 |
-
<li><strong>5-Panel View:</strong> Complete analysis with heatmap</li>
|
366 |
</ul>
|
367 |
</div>
|
368 |
""")
|
369 |
|
370 |
with gr.Column(scale=2):
|
371 |
-
gr.Markdown("###
|
372 |
|
373 |
output_image = gr.Image(
|
374 |
-
label="Your Attention U-Net Analysis
|
375 |
type="pil",
|
376 |
height=500
|
377 |
)
|
378 |
|
379 |
analysis_output = gr.Markdown(
|
380 |
-
value="Upload a brain MRI image to test YOUR trained Attention U-Net model
|
381 |
elem_id="analysis"
|
382 |
)
|
383 |
-
|
384 |
-
# Footer highlighting your model with heatmap features
|
385 |
gr.HTML("""
|
386 |
<div style="margin-top: 30px; padding: 25px; background-color: #F8FAFC; border-radius: 15px; border: 2px solid #8B5CF6;">
|
387 |
<div style="display: grid; grid-template-columns: 1fr 1fr; gap: 30px;">
|
388 |
<div>
|
389 |
-
<h4 style="color: #8B5CF6; margin-bottom: 15px;"
|
390 |
<p><strong>Architecture:</strong> Attention U-Net with skip connections</p>
|
391 |
<p><strong>Performance:</strong> Dice: 0.8420, IoU: 0.7297, Accuracy: 98.90%</p>
|
392 |
<p><strong>Training:</strong> Your own dataset-specific training</p>
|
393 |
<p><strong>Features:</strong> [32, 64, 128, 256] channel progression</p>
|
394 |
-
<p><strong>NEW:</strong> Continuous heatmap visualization for confidence</p>
|
395 |
</div>
|
396 |
<div>
|
397 |
-
<h4 style="color: #DC2626; margin-bottom: 15px;"
|
398 |
<p style="color: #DC2626; font-weight: 600; line-height: 1.4;">
|
399 |
This is YOUR personally trained AI model for <strong>research purposes only</strong>.<br>
|
400 |
Results reflect your model's training performance.<br>
|
@@ -404,7 +399,7 @@ with gr.Blocks(css=css, title="🧠 Your Attention U-Net Model with Heatmap", th
|
|
404 |
</div>
|
405 |
<hr style="margin: 20px 0; border: none; border-top: 2px solid #E5E7EB;">
|
406 |
<p style="text-align: center; color: #6B7280; margin: 10px 0; font-weight: 600;">
|
407 |
-
|
408 |
</p>
|
409 |
</div>
|
410 |
""")
|
@@ -412,27 +407,32 @@ with gr.Blocks(css=css, title="🧠 Your Attention U-Net Model with Heatmap", th
|
|
412 |
# Event handlers
|
413 |
analyze_btn.click(
|
414 |
fn=predict_tumor,
|
415 |
-
inputs=[image_input],
|
416 |
outputs=[output_image, analysis_output],
|
417 |
show_progress=True
|
418 |
)
|
419 |
|
|
|
|
|
|
|
|
|
|
|
|
|
420 |
clear_btn.click(
|
421 |
fn=clear_all,
|
422 |
inputs=[],
|
423 |
-
outputs=[image_input, output_image, analysis_output]
|
424 |
)
|
425 |
|
426 |
if __name__ == "__main__":
|
427 |
-
print("
|
428 |
-
print("
|
429 |
-
print("
|
430 |
-
print("
|
431 |
-
print("🔥 NEW: Heatmap visualization added!")
|
432 |
|
433 |
app.launch(
|
434 |
server_name="0.0.0.0",
|
435 |
server_port=7860,
|
436 |
show_error=True,
|
437 |
share=False
|
438 |
-
)
|
|
|
10 |
import torchvision.transforms.functional as TF
|
11 |
import urllib.request
|
12 |
import os
|
13 |
+
import random
|
14 |
+
import kagglehub
|
15 |
|
16 |
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
17 |
model = None
|
18 |
|
19 |
+
# Download dataset
|
20 |
+
dataset_path = kagglehub.dataset_download('nikhilroxtomar/brain-tumor-segmentation')
|
21 |
+
image_path = os.path.join(dataset_path, 'images')
|
22 |
+
mask_path = os.path.join(dataset_path, 'masks')
|
23 |
+
test_imgs = sorted([f for f in os.listdir(image_path) if f.endswith('.jpg') or f.endswith('.png')])
|
24 |
+
test_masks = sorted([f for f in os.listdir(mask_path) if f.endswith('.jpg') or f.endswith('.png')])
|
25 |
+
|
26 |
# Define your Attention U-Net architecture (from your training code)
|
27 |
class DoubleConv(nn.Module):
|
28 |
def __init__(self, in_channels, out_channels):
|
|
|
74 |
self.downs = nn.ModuleList()
|
75 |
self.attentions = nn.ModuleList()
|
76 |
self.pool = nn.MaxPool2d(kernel_size=2, stride=2)
|
|
|
77 |
# Down part of UNET
|
78 |
for feature in features:
|
79 |
self.downs.append(DoubleConv(in_channels, feature))
|
80 |
in_channels = feature
|
|
|
81 |
# Bottleneck
|
82 |
self.bottleneck = DoubleConv(features[-1], features[-1]*2)
|
83 |
|
|
|
86 |
self.ups.append(nn.ConvTranspose2d(feature*2, feature, kernel_size=2, stride=2))
|
87 |
self.attentions.append(AttentionBlock(F_g=feature, F_l=feature, F_int=feature // 2))
|
88 |
self.ups.append(DoubleConv(feature*2, feature))
|
|
|
89 |
self.final_conv = nn.Conv2d(features[0], out_channels, kernel_size=1)
|
|
|
90 |
def forward(self, x):
|
91 |
skip_connections = []
|
92 |
for down in self.downs:
|
93 |
x = down(x)
|
94 |
skip_connections.append(x)
|
95 |
x = self.pool(x)
|
|
|
96 |
x = self.bottleneck(x)
|
97 |
skip_connections = skip_connections[::-1] #reverse list
|
|
|
98 |
for idx in range(0, len(self.ups), 2): #do up and double_conv
|
99 |
x = self.ups[idx](x)
|
100 |
skip_connection = skip_connections[idx//2]
|
|
|
103 |
skip_connection = self.attentions[idx // 2](skip_connection, x)
|
104 |
concat_skip = torch.cat((skip_connection, x), dim=1)
|
105 |
x = self.ups[idx+1](concat_skip)
|
|
|
106 |
return self.final_conv(x)
|
107 |
|
108 |
def download_model():
|
|
|
111 |
model_path = "best_attention_model.pth.tar"
|
112 |
|
113 |
if not os.path.exists(model_path):
|
114 |
+
print("Downloading your trained model...")
|
115 |
try:
|
116 |
urllib.request.urlretrieve(model_url, model_path)
|
117 |
+
print("Model downloaded successfully!")
|
118 |
except Exception as e:
|
119 |
+
print(f"Failed to download model: {e}")
|
120 |
return None
|
121 |
else:
|
122 |
+
print("Model already exists!")
|
123 |
|
124 |
return model_path
|
125 |
|
|
|
128 |
global model
|
129 |
if model is None:
|
130 |
try:
|
131 |
+
print("Loading your trained Attention U-Net model...")
|
132 |
|
133 |
# Download model if needed
|
134 |
model_path = download_model()
|
|
|
143 |
model.load_state_dict(checkpoint["state_dict"])
|
144 |
model.eval()
|
145 |
|
146 |
+
print("Your Attention U-Net model loaded successfully!")
|
147 |
except Exception as e:
|
148 |
+
print(f"Error loading your model: {e}")
|
149 |
model = None
|
150 |
return model
|
151 |
|
|
|
163 |
|
164 |
return val_test_transform(image).unsqueeze(0) # Add batch dimension
|
165 |
|
166 |
+
def predict_tumor(image, mask=None):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
167 |
current_model = load_your_attention_model()
|
168 |
|
169 |
if current_model is None:
|
170 |
+
return None, "Failed to load your trained model."
|
171 |
if image is None:
|
172 |
+
return None, "Please upload an image first."
|
173 |
|
174 |
try:
|
175 |
+
print("Processing with YOUR trained Attention U-Net...")
|
176 |
|
177 |
# Use the exact preprocessing from your Colab code
|
178 |
input_tensor = preprocess_for_your_model(image).to(device)
|
179 |
|
180 |
# Predict using your model (exactly like your Colab code)
|
181 |
with torch.no_grad():
|
182 |
+
pred_mask = torch.sigmoid(current_model(input_tensor))
|
183 |
+
pred_mask_binary = (pred_mask > 0.5).float()
|
184 |
|
185 |
+
# Convert to numpy (like your Colab code)
|
186 |
pred_mask_np = pred_mask_binary.cpu().squeeze().numpy()
|
187 |
+
prob_mask_np = pred_mask.cpu().squeeze().numpy() # Probability for heatmap
|
188 |
original_np = np.array(image.convert('L').resize((256, 256)))
|
189 |
|
190 |
+
# Create inverted mask for visualization (like your Colab code)
|
191 |
inv_pred_mask_np = np.where(pred_mask_np == 1, 0, 255)
|
192 |
|
193 |
+
# Create tumor-only image (like your Colab code)
|
194 |
tumor_only = np.where(pred_mask_np == 1, original_np, 255)
|
195 |
|
196 |
+
# Handle ground truth if provided
|
197 |
+
mask_np = None
|
198 |
+
dice_score = None
|
199 |
+
iou_score = None
|
200 |
+
if mask is not None:
|
201 |
+
mask_transform = transforms.Compose([
|
202 |
+
transforms.Resize((256,256)),
|
203 |
+
transforms.ToTensor()
|
204 |
+
])
|
205 |
+
mask_tensor = mask_transform(mask).squeeze().numpy()
|
206 |
+
mask_np = (mask_tensor > 0.5).astype(float)
|
207 |
+
|
208 |
+
intersection = np.logical_and(pred_mask_np, mask_np).sum()
|
209 |
+
union = np.logical_or(pred_mask_np, mask_np).sum()
|
210 |
+
iou_score = intersection / (union + 1e-7)
|
211 |
+
dice_score = (2 * intersection) / (pred_mask_np.sum() + mask_np.sum() + 1e-7)
|
212 |
|
213 |
+
# Create visualization (5-panel layout)
|
214 |
fig, axes = plt.subplots(1, 5, figsize=(25, 5))
|
215 |
+
fig.suptitle('Your Attention U-Net Results', fontsize=16, fontweight='bold')
|
216 |
|
217 |
+
titles = ["Original Image", "Ground Truth", "Predicted Mask", "Tumor Only", "Heatmap"]
|
218 |
+
images = [original_np, mask_np if mask_np is not None else np.zeros_like(original_np), inv_pred_mask_np, tumor_only, prob_mask_np]
|
219 |
+
cmaps = ['gray', 'gray', 'gray', 'gray', 'hot']
|
220 |
|
221 |
for i, ax in enumerate(axes):
|
222 |
+
ax.imshow(images[i], cmap=cmaps[i])
|
|
|
|
|
|
|
223 |
ax.set_title(titles[i], fontsize=12, fontweight='bold')
|
224 |
ax.axis('off')
|
225 |
|
|
|
233 |
|
234 |
result_image = Image.open(buf)
|
235 |
|
236 |
+
# Calculate statistics (like your Colab code)
|
237 |
tumor_pixels = np.sum(pred_mask_np)
|
238 |
total_pixels = pred_mask_np.size
|
239 |
tumor_percentage = (tumor_pixels / total_pixels) * 100
|
240 |
|
241 |
# Calculate confidence metrics
|
242 |
+
max_confidence = torch.max(pred_mask).item()
|
243 |
+
mean_confidence = torch.mean(pred_mask).item()
|
244 |
|
245 |
analysis_text = f"""
|
246 |
+
## Your Attention U-Net Analysis Results
|
247 |
+
### Detection Summary:
|
248 |
+
- **Status**: {'TUMOR DETECTED' if tumor_pixels > 50 else 'NO SIGNIFICANT TUMOR'}
|
249 |
- **Tumor Area**: {tumor_percentage:.2f}% of brain region
|
250 |
- **Tumor Pixels**: {tumor_pixels:,} pixels
|
251 |
- **Max Confidence**: {max_confidence:.4f}
|
252 |
- **Mean Confidence**: {mean_confidence:.4f}
|
253 |
+
"""
|
254 |
+
if dice_score is not None and iou_score is not None:
|
255 |
+
analysis_text += f"""
|
256 |
+
- **Dice Score**: {dice_score:.4f}
|
257 |
+
- **IoU Score**: {iou_score:.4f}
|
258 |
+
"""
|
259 |
+
analysis_text += """
|
260 |
+
### Model Information:
|
261 |
- **Architecture**: YOUR trained Attention U-Net
|
262 |
- **Training Performance**: Dice: 0.8420, IoU: 0.7297
|
263 |
- **Input**: Grayscale (single channel)
|
264 |
+
- **Output**: Binary segmentation mask
|
265 |
- **Device**: {device.type.upper()}
|
266 |
+
### Model Performance:
|
|
|
267 |
- **Training Accuracy**: 98.90%
|
268 |
- **Best Dice Score**: 0.8420
|
269 |
- **Best IoU Score**: 0.7297
|
270 |
- **Training Dataset**: Brain tumor segmentation dataset
|
271 |
+
### Processing Details:
|
|
|
272 |
- **Preprocessing**: Resize(256×256) + ToTensor (your exact method)
|
273 |
- **Threshold**: 0.5 (sigmoid > 0.5)
|
274 |
- **Architecture**: Attention gates + Skip connections
|
275 |
- **Features**: [32, 64, 128, 256] channels
|
276 |
+
### Medical Disclaimer:
|
|
|
|
|
277 |
This is YOUR trained AI model for **research and educational purposes only**.
|
278 |
Results should be validated by medical professionals. Not for clinical diagnosis.
|
279 |
+
### Model Quality:
|
280 |
+
This is your own trained model with {tumor_percentage:.2f}% detection capability!
|
|
|
281 |
"""
|
282 |
|
283 |
+
print(f"Your model analysis completed! Tumor area: {tumor_percentage:.2f}%")
|
284 |
return result_image, analysis_text
|
285 |
|
286 |
except Exception as e:
|
287 |
+
error_msg = f"Error with your model: {str(e)}"
|
288 |
print(error_msg)
|
289 |
return None, error_msg
|
290 |
|
291 |
+
def load_random_sample():
|
292 |
+
if not test_imgs:
|
293 |
+
return None, None, "Dataset not available."
|
294 |
+
rand_idx = random.randint(0, len(test_imgs) - 1)
|
295 |
+
img_path = os.path.join(image_path, test_imgs[rand_idx])
|
296 |
+
msk_path = os.path.join(mask_path, test_masks[rand_idx]) # Assuming paired by index
|
297 |
+
image = Image.open(img_path).convert('L')
|
298 |
+
mask = Image.open(msk_path).convert('L')
|
299 |
+
return image, mask, "Loaded random sample from dataset."
|
300 |
+
|
301 |
def clear_all():
|
302 |
+
return None, None, "Upload a brain MRI image to test YOUR trained Attention U-Net model", None
|
303 |
|
304 |
# Enhanced CSS for your model
|
305 |
css = """
|
|
|
319 |
"""
|
320 |
|
321 |
# Create Gradio interface for your model
|
322 |
+
with gr.Blocks(css=css, title="Your Attention U-Net Model", theme=gr.themes.Soft()) as app:
|
323 |
|
324 |
gr.HTML("""
|
325 |
<div id="title">
|
326 |
+
<h1>Your Attention U-Net Model</h1>
|
327 |
<p style="font-size: 18px; margin-top: 15px;">
|
328 |
+
Using Your Own Trained Model • Dice: 0.8420 • IoU: 0.7297
|
329 |
</p>
|
330 |
<p style="font-size: 14px; margin-top: 10px; opacity: 0.9;">
|
331 |
Loaded from: ArchCoder/the-op-segmenter HuggingFace Space
|
|
|
333 |
</div>
|
334 |
""")
|
335 |
|
336 |
+
mask_state = gr.State(None)
|
337 |
+
|
338 |
with gr.Row():
|
339 |
with gr.Column(scale=1):
|
340 |
+
gr.Markdown("### Upload Brain MRI")
|
341 |
|
342 |
image_input = gr.Image(
|
343 |
label="Brain MRI Scan",
|
|
|
347 |
)
|
348 |
|
349 |
with gr.Row():
|
350 |
+
analyze_btn = gr.Button("Analyze with YOUR Model", variant="primary", scale=1, size="lg")
|
351 |
+
random_btn = gr.Button("Load Random Sample", variant="secondary", scale=1, size="lg")
|
352 |
+
clear_btn = gr.Button("Clear", variant="secondary", scale=1)
|
353 |
|
354 |
gr.HTML("""
|
355 |
<div style="margin-top: 20px; padding: 20px; background: linear-gradient(135deg, #F3E8FF 0%, #EDE9FE 100%); border-radius: 10px; border-left: 4px solid #8B5CF6;">
|
356 |
+
<h4 style="color: #8B5CF6; margin-bottom: 15px;">Your Model Features:</h4>
|
357 |
<ul style="margin: 10px 0; padding-left: 20px; line-height: 1.6;">
|
358 |
<li><strong>Personal Model:</strong> Your own trained Attention U-Net</li>
|
359 |
<li><strong>Proven Performance:</strong> 84.2% Dice Score, 72.97% IoU</li>
|
360 |
<li><strong>Attention Gates:</strong> Advanced feature selection</li>
|
361 |
<li><strong>Clean Output:</strong> Binary segmentation masks</li>
|
362 |
+
<li><strong>5-Panel View:</strong> Complete analysis like your Colab</li>
|
|
|
363 |
</ul>
|
364 |
</div>
|
365 |
""")
|
366 |
|
367 |
with gr.Column(scale=2):
|
368 |
+
gr.Markdown("### Your Model Results")
|
369 |
|
370 |
output_image = gr.Image(
|
371 |
+
label="Your Attention U-Net Analysis",
|
372 |
type="pil",
|
373 |
height=500
|
374 |
)
|
375 |
|
376 |
analysis_output = gr.Markdown(
|
377 |
+
value="Upload a brain MRI image to test YOUR trained Attention U-Net model.",
|
378 |
elem_id="analysis"
|
379 |
)
|
380 |
+
# Footer highlighting your model
|
|
|
381 |
gr.HTML("""
|
382 |
<div style="margin-top: 30px; padding: 25px; background-color: #F8FAFC; border-radius: 15px; border: 2px solid #8B5CF6;">
|
383 |
<div style="display: grid; grid-template-columns: 1fr 1fr; gap: 30px;">
|
384 |
<div>
|
385 |
+
<h4 style="color: #8B5CF6; margin-bottom: 15px;">Your Personal AI Model</h4>
|
386 |
<p><strong>Architecture:</strong> Attention U-Net with skip connections</p>
|
387 |
<p><strong>Performance:</strong> Dice: 0.8420, IoU: 0.7297, Accuracy: 98.90%</p>
|
388 |
<p><strong>Training:</strong> Your own dataset-specific training</p>
|
389 |
<p><strong>Features:</strong> [32, 64, 128, 256] channel progression</p>
|
|
|
390 |
</div>
|
391 |
<div>
|
392 |
+
<h4 style="color: #DC2626; margin-bottom: 15px;">Your Model Disclaimer</h4>
|
393 |
<p style="color: #DC2626; font-weight: 600; line-height: 1.4;">
|
394 |
This is YOUR personally trained AI model for <strong>research purposes only</strong>.<br>
|
395 |
Results reflect your model's training performance.<br>
|
|
|
399 |
</div>
|
400 |
<hr style="margin: 20px 0; border: none; border-top: 2px solid #E5E7EB;">
|
401 |
<p style="text-align: center; color: #6B7280; margin: 10px 0; font-weight: 600;">
|
402 |
+
Your Personal Attention U-Net • Downloaded from HuggingFace • Research-Grade Performance
|
403 |
</p>
|
404 |
</div>
|
405 |
""")
|
|
|
407 |
# Event handlers
|
408 |
analyze_btn.click(
|
409 |
fn=predict_tumor,
|
410 |
+
inputs=[image_input, mask_state],
|
411 |
outputs=[output_image, analysis_output],
|
412 |
show_progress=True
|
413 |
)
|
414 |
|
415 |
+
random_btn.click(
|
416 |
+
fn=load_random_sample,
|
417 |
+
inputs=[],
|
418 |
+
outputs=[image_input, mask_state, analysis_output]
|
419 |
+
)
|
420 |
+
|
421 |
clear_btn.click(
|
422 |
fn=clear_all,
|
423 |
inputs=[],
|
424 |
+
outputs=[image_input, output_image, analysis_output, mask_state]
|
425 |
)
|
426 |
|
427 |
if __name__ == "__main__":
|
428 |
+
print("Starting YOUR Attention U-Net Model System...")
|
429 |
+
print("Using your personally trained model")
|
430 |
+
print("Auto-downloading from HuggingFace...")
|
431 |
+
print("Expected performance: Dice 0.8420, IoU 0.7297")
|
|
|
432 |
|
433 |
app.launch(
|
434 |
server_name="0.0.0.0",
|
435 |
server_port=7860,
|
436 |
show_error=True,
|
437 |
share=False
|
438 |
+
)
|