Upload 4 files

.gitattributes

@@ -0,0 +1 @@
+CC_net[[:space:]](1).pt filter=lfs diff=lfs merge=lfs -text

CC_net (1).pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b61ad39bb8f2872cff371265b3ad4ecbf9c5a201d64225f92d6bcc937d9e112b
+size 95648689

ResNet_for_CC.py

@@ -0,0 +1,93 @@
+import torch
+import torch.nn as nn
+import torchvision.models as models
+
+class ResClassifier(nn.Module):
+    """
+    A classifier with two fully connected layers followed by a final linear layer.
+    Uses BatchNorm, ReLU activations, and Dropout for better generalization.
+    """
+    def __init__(self, num_classes=14):
+        super(ResClassifier, self).__init__()
+        
+        # First fully connected layer: reduces 128D features to 64D
+        self.fc1 = nn.Sequential(
+            nn.Linear(128, 64),
+            nn.BatchNorm1d(64, affine=True),
+            nn.ReLU(inplace=True),
+            nn.Dropout()
+        )
+        
+        # Second fully connected layer: retains 64D features
+        self.fc2 = nn.Sequential(
+            nn.Linear(64, 64),
+            nn.BatchNorm1d(64, affine=True),
+            nn.ReLU(inplace=True),
+            nn.Dropout()
+        )
+        
+        # Final classification layer mapping 64D features to class logits
+        self.fc3 = nn.Linear(64, num_classes)
+
+    def forward(self, x):
+        """
+        Forward pass through the classifier.
+        Returns class logits after two hidden layers.
+        """
+        x = self.fc1(x)  # First FC layer
+        x = self.fc2(x)  # Second FC layer
+        output = self.fc3(x)  # Final classification layer
+        return output
+
+
+class CC_model(nn.Module):
+    """
+    Clothing Classification Model based on ResNet50.
+    Extracts deep features and uses two independent classifiers for predictions.
+    """
+    def __init__(self, num_classes1=14, num_classes2=None):
+        super(CC_model, self).__init__()
+        
+        # If num_classes2 is not specified, default to num_classes1
+        num_classes2 = num_classes2 if num_classes2 else num_classes1
+        assert num_classes1 == num_classes2  # Ensure both classifiers predict the same categories
+        
+        self.num_classes = num_classes1
+        
+        # Load a pretrained ResNet-50 model as the feature extractor
+        self.model_resnet = models.resnet50(weights='ResNet50_Weights.DEFAULT')
+        
+        # Remove ResNet's original classification layer to use as a feature extractor
+        num_ftrs = self.model_resnet.fc.in_features
+        self.model_resnet.fc = nn.Identity()  # Identity layer keeps feature dimensions
+
+        # Additional transformation layer reducing feature size to 128D
+        self.dr = nn.Linear(num_ftrs, 128)
+
+        # Two independent classifiers
+        self.fc1 = ResClassifier(num_classes1)
+        self.fc2 = ResClassifier(num_classes1)
+
+    def forward(self, x, detach_feature=False):
+        """
+        Forward pass through the model.
+        Extracts deep features from ResNet and processes them through classifiers.
+        """
+        with torch.no_grad():
+            # Extract deep features using ResNet-50 (without its original classification head)
+            feature = self.model_resnet(x)
+
+        # Generate transformed features (128D) using the custom linear layer
+        dr_feature = self.dr(feature)
+
+        if detach_feature:
+            dr_feature = dr_feature.detach()  # Detach feature for non-trainable forward pass
+
+        # Pass features through two independent classifiers
+        out1 = self.fc1(dr_feature)
+        out2 = self.fc2(dr_feature)
+
+        # Compute the mean prediction from both classifiers
+        output_mean = (out1 + out2) / 2
+
+        return dr_feature, output_mean  # Returning feature embeddings and final prediction

app (1).py

@@ -0,0 +1,95 @@
+import gradio as gr
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torchvision.transforms as transforms
+from PIL import Image
+from ResNet_for_CC import CC_model  # Import the model
+
+# Set device (CPU/GPU)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# Load the trained CC_model
+model_path = "CC_net.pt"
+model = CC_model(num_classes1=14)
+
+# Load model weights
+state_dict = torch.load(model_path, map_location=device)
+model.load_state_dict(state_dict, strict=False)
+model.to(device)
+model.eval()
+
+# Clothing1M Class Labels
+class_labels = [
+    "T-Shirt", "Shirt", "Knitwear", "Chiffon", "Sweater", "Hoodie",
+    "Windbreaker", "Jacket", "Downcoat", "Suit", "Shawl", "Dress",
+    "Vest", "Underwear"
+]
+
+# ✅ **Updated Image Preprocessing Function**
+def preprocess_image(image):
+    """Applies necessary transformations to the input image."""
+    transform = transforms.Compose([
+        transforms.Resize(256),
+        transforms.CenterCrop(224),
+        transforms.ToTensor(),
+        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+    ])
+    return transform(image).unsqueeze(0).to(device)
+
+# ✅ **Classification Function**
+def classify_image(image):
+    """Processes the input image and returns the predicted clothing category."""
+    print("\n[INFO] Received image for classification.")
+
+    try:
+        image = Image.fromarray(image)  # Ensure conversion to PIL format
+        image = preprocess_image(image)  # Apply transformations
+        print("[INFO] Image transformed and moved to device.")
+
+        with torch.no_grad():
+            output = model(image)
+            
+            # ✅ Ensure output is a tensor (handle tuple case)
+            if isinstance(output, tuple):
+                output = output[1]  # Extract the actual output tensor
+
+            print(f"[DEBUG] Model output shape: {output.shape}")
+            print(f"[DEBUG] Model output values: {output}")
+
+            if output.shape[1] != 14:
+                return f"[ERROR] Model output mismatch! Expected 14 but got {output.shape[1]}."
+
+            # Convert logits to probabilities
+            probabilities = F.softmax(output, dim=1)
+            print(f"[DEBUG] Softmax probabilities: {probabilities}")
+
+            # Get predicted class index
+            predicted_class = torch.argmax(probabilities, dim=1).item()
+            print(f"[INFO] Predicted class index: {predicted_class} (Class: {class_labels[predicted_class]})")
+
+            # Validate and return the prediction
+            if 0 <= predicted_class < len(class_labels):
+                predicted_label = class_labels[predicted_class]
+                confidence = probabilities[0][predicted_class].item() * 100
+                return f"Predicted Class: {predicted_label} (Confidence: {confidence:.2f}%)"
+            else:
+                return "[ERROR] Model returned an invalid class index."
+
+    except Exception as e:
+        print(f"[ERROR] Exception during classification: {e}")
+        return "Error in classification. Check console for details."
+
+# ✅ **Gradio Interface**
+interface = gr.Interface(
+    fn=classify_image,
+    inputs=gr.Image(type="numpy"),
+    outputs="text",
+    title="Clothing1M Image Classifier",
+    description="Upload a clothing image, and the model will classify it into one of the 14 categories."
+)
+
+# ✅ **Run the Interface**
+if __name__ == "__main__":
+    print("[INFO] Launching Gradio interface...")
+    interface.launch()

requirements.txt

@@ -0,0 +1,7 @@
+clip==0.2.0
+numpy==1.23.4
+openai_clip==1.0.1
+Pillow==9.4.0
+torch==2.6.0
+torchvision==0.21.0
+tqdm==4.64.1