Update app.py

app.py

@@ -22,6 +22,9 @@ def predict_caption(image):
     if image.mode != "RGB":
         image = image.convert(mode="RGB")
 
+    # Resize image to a fixed size (ViT typically requires 224x224 or 384x384)
+    image = image.resize((384, 384))  # Resize to 384x384 for ViT
+
     # Process image with padding enabled to handle batched tensor conversion
     pixel_values = feature_extractor(images=[image], return_tensors="pt", padding=True).pixel_values.to(device)