Update app.py

app.py

@@ -31,17 +31,17 @@ async def predict(files: List[UploadFile] = File(...)):
             image = Image.open(io.BytesIO(image_data)).convert("RGB")
             
             # Generate description
-            inputs = processor(text=["Describe the image in detail."], images=[image], return_tensors="pt")
+            inputs = processor(text=["<image> Describe the image in detail."], images=[image], return_tensors="pt")
             outputs = model.generate(**inputs, max_length=100)
             description = processor.decode(outputs[0], skip_special_tokens=True).replace("Describe the image in detail.", "").strip()
             
             # Extract signs/number plates (OCR)
-            inputs_ocr = processor(text=["Extract all text visible in the image."], images=[image], return_tensors="pt")
+            inputs_ocr = processor(text=["<image> Extract all text visible in the image."], images=[image], return_tensors="pt")
             ocr_outputs = model.generate(**inputs_ocr, max_length=100)
             signs_text = processor.decode(ocr_outputs[0], skip_special_tokens=True).replace("Extract all text visible in the image.", "").strip()
             
             # Detect harmful objects/blood
-            inputs_detect = processor(text=["List all objects in the image."], images=[image], return_tensors="pt")
+            inputs_detect = processor(text=["<image> List all objects in the image."], images=[image], return_tensors="pt")
             detect_outputs = model.generate(**inputs_detect, max_length=100)
             detected_objects = processor.decode(detect_outputs[0], skip_special_tokens=True).lower()
             harmful_detected = any(obj in detected_objects for obj in harmful_objects) and "Detected: " + ", ".join([obj for obj in harmful_objects if obj in detected_objects]) or "None detected"
@@ -89,17 +89,17 @@ def gradio_predict(*images):
             image = Image.fromarray(image).convert("RGB")
             
             # Generate description
-            inputs = processor(text=["Describe the image in detail."], images=[image], return_tensors="pt")
+            inputs = processor(text=["<image> Describe the image in detail."], images=[image], return_tensors="pt")
             outputs = model.generate(**inputs, max_length=100)
             description = processor.decode(outputs[0], skip_special_tokens=True).replace("Describe the image in detail.", "").strip()
             
             # Extract signs/number plates (OCR)
-            inputs_ocr = processor(text=["Extract all text visible in the image."], images=[image], return_tensors="pt")
+            inputs_ocr = processor(text=["<image> Extract all text visible in the image."], images=[image], return_tensors="pt")
             ocr_outputs = model.generate(**inputs_ocr, max_length=100)
             signs_text = processor.decode(ocr_outputs[0], skip_special_tokens=True).replace("Extract all text visible in the image.", "").strip()
             
             # Detect harmful objects/blood
-            inputs_detect = processor(text=["List all objects in the image."], images=[image], return_tensors="pt")
+            inputs_detect = processor(text=["<image> List all objects in the image."], images=[image], return_tensors="pt")
             detect_outputs = model.generate(**inputs_detect, max_length=100)
             detected_objects = processor.decode(detect_outputs[0], skip_special_tokens=True).lower()
             harmful_detected = any(obj in detected_objects for obj in harmful_objects) and "Detected: " + ", ".join([obj for obj in harmful_objects if obj in detected_objects]) or "None detected"