Update app.py

app.py

@@ -1,31 +1,54 @@
 import torch
-from PIL import Image
+import torch.nn as nn
 import torchvision.transforms as transforms
+from PIL import Image
+import numpy as np
 import gradio as gr
 
 # -------------------------------
-# 1️⃣ تحميل النموذج المحفوظ كاملًا
+# 1️⃣ إعداد الفئات
 # -------------------------------
-# تأكدي أن best_model_2.pth تم حفظه باستخدام torch.save(model, "best_model_2.pth")
-model = torch.load("best_model_2.pth", map_location=torch.device('cpu'))
-model.eval()
+CLASSES = ["NONE", "INFECTION", "ISCHAEMIA", "BOTH"]
 
 # -------------------------------
-# 2️⃣ إعداد الفئات
+# 2️⃣ تعريف نموذج DenseShuffleGCANet
 # -------------------------------
-CLASSES = ["NONE", "INFECTION", "ISCHAEMIA", "BOTH"]
+class DenseShuffleGCANet(nn.Module):
+    def __init__(self, num_classes=4, handcrafted_feature_dim=49):
+        super(DenseShuffleGCANet, self).__init__()
+        # مثال على backbone، عدلي حسب الكود الأصلي
+        self.backbone = nn.Sequential(
+            nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1),
+            nn.ReLU(),
+            nn.AdaptiveAvgPool2d((1,1))
+        )
+        self.fc_handcrafted = nn.Linear(handcrafted_feature_dim, 32)
+        self.classifier = nn.Linear(64 + 32, num_classes)
+    
+    def forward(self, x_image, x_features):
+        x_img = self.backbone(x_image)
+        x_img = x_img.view(x_img.size(0), -1)
+        x_feat = self.fc_handcrafted(x_features)
+        x = torch.cat([x_img, x_feat], dim=1)
+        out = self.classifier(x)
+        return out
 
 # -------------------------------
-# 3️⃣ دالة استخراج الخصائص اليدوية
+# 3️⃣ تحميل النموذج مع weights
+# -------------------------------
+model = DenseShuffleGCANet(num_classes=4, handcrafted_feature_dim=49)
+model.load_state_dict(torch.load("best_model_2.pth", map_location=torch.device('cpu')))
+model.eval()
+
+# -------------------------------
+# 4️⃣ دالة استخراج الخصائص اليدوية
 # -------------------------------
 def extract_handcrafted_features(image_array):
-    # مثال عشوائي لتوضيح الفكرة، عدلي حسب ما كنتِ تستخدمينه
-    import numpy as np
-    features = np.random.rand(49).astype(np.float32)
+    features = np.random.rand(49).astype(np.float32)  # عدلي حسب خصائصك الحقيقية
     return torch.tensor(features)
 
 # -------------------------------
-# 4️⃣ دالة التنبؤ
+# 5️⃣ دالة التنبؤ
 # -------------------------------
 def predict_image(image: Image.Image):
     transform = transforms.Compose([
@@ -34,8 +57,6 @@ def predict_image(image: Image.Image):
         transforms.Normalize(mean=[0.485,0.456,0.406], std=[0.229,0.224,0.225])
     ])
     image_tensor = transform(image).unsqueeze(0)
-
-    # استخراج الخصائص اليدوية
     features = extract_handcrafted_features(np.array(image)).unsqueeze(0)
 
     with torch.no_grad():
@@ -46,7 +67,7 @@ def predict_image(image: Image.Image):
     return {CLASSES[i]: float(probs[i]) for i in range(len(CLASSES))}, pred_class
 
 # -------------------------------
-# 5️⃣ واجهة Gradio
+# 6️⃣ واجهة Gradio
 # -------------------------------
 interface = gr.Interface(
     fn=predict_image,
@@ -58,3 +79,4 @@ interface = gr.Interface(
 
 if __name__ == "__main__":
     interface.launch()
+