Approach 2 is working

app.py

@@ -1,28 +1,35 @@
 import torch
-from transformers import AutoModel, AutoImageProcessor
+from transformers import CLIPProcessor, CLIPModel
 from PIL import Image
 import gradio as gr
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
-# Load DINOv2 model
-model_name = "facebook/dinov2-base"
-model = AutoModel.from_pretrained(model_name).to(device)
-processor = AutoImageProcessor.from_pretrained(model_name)
+# Load the model
+model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32").to(device)
+processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
 
-# Embed a normal sample (you can do this during inference without storing)
-# You can even hardcode embeddings later if you want
+# Load normal image for reference
+normal_image = Image.open("normal_sample.jpg")
+
+with torch.no_grad():
+    inputs = processor(images=normal_image, return_tensors="pt").to(device)
+    normal_features = model.get_image_features(**inputs)
+    normal_features = normal_features / normal_features.norm(p=2, dim=-1, keepdim=True)
 
 def detect_anomaly(img):
-    inputs = processor(images=img, return_tensors="pt").to(device)
     with torch.no_grad():
-        features = model(**inputs).last_hidden_state.mean(dim=1)
-    
-    # Fake logic for now (just show feature norm)
-    # Later you can compare with normal samples' embeddings
-    score = torch.norm(features).item()
-    
-    return f"Feature Norm (use for anomaly logic): {score:.2f}"
+        inputs = processor(images=img, return_tensors="pt").to(device)
+        test_features = model.get_image_features(**inputs)
+        test_features = test_features / test_features.norm(p=2, dim=-1, keepdim=True)
+        
+        similarity = (test_features @ normal_features.T).item()
+        
+    if similarity < 0.8:  # threshold example
+        result = "Anomaly Detected"
+    else:
+        result = "Normal"
+    return f"Similarity: {similarity:.2f} | {result}"
 
 gr.Interface(fn=detect_anomaly,
              inputs=gr.Image(),