Upload folder using huggingface_hub

app/services/vision_service.py

@@ -1,36 +1,65 @@
 import logging
+import time
+import traceback
 from gradio_client import Client, handle_file
 from app.core.config import settings
 
 class VisionService:
     def __init__(self):
         self.client = None
+        self.endpoint = "MinhDS/Florence-2-Demo"
         try:
-            self.client = Client("MinhDS/Florence-2-Demo")
+            logging.info(f"[VisionService] Initializing client for {self.endpoint}...")
+            self.client = Client(self.endpoint)
+            logging.info(f"[VisionService] Client initialized successfully.")
         except Exception as e:
-            logging.error(f"Failed to initialize Florence-2 client: {e}")
+            logging.error(f"[VisionService] Failed to initialize client: {e}")
+            logging.error(traceback.format_exc())
 
     async def describe_screenshot(self, image_path: str):
         if not self.client:
+            logging.warning("[VisionService] describe_screenshot called but client is not available.")
             return "Vision service not available."
 
+        task_prompt = "More Detailed Caption"
+        model_id = "microsoft/Florence-2-base"
+        
         try:
-            logging.info(f"Calling Florence-2 vision API for {image_path}...")
-            # Using 'More Detailed Caption' for rich textual description
-            # We use a long timeout as vision tasks can be slow
+            logging.info(f"[VisionService] Starting description for {image_path}...")
+            logging.info(f"[VisionService] Task: {task_prompt}, Model: {model_id}")
+            
+            start_time = time.time()
+            
+            # Note: client.predict is synchronous in gradio_client, 
+            # but we are in an async function. For now keeping it simple as it was.
             result = self.client.predict(
                 image=handle_file(image_path),
-                task_prompt="More Detailed Caption",
+                task_prompt=task_prompt,
                 text_input=None,
-                model_id="microsoft/Florence-2-base",
+                model_id=model_id,
                 api_name="/process_image"
             )
-            # result[0] is the text output, result[1] is the output image (which we ignore)
+            
+            duration = time.time() - start_time
+            logging.info(f"[VisionService] API call completed in {duration:.2f} seconds.")
+            
+            # Log result type and structure for debugging
+            logging.info(f"[VisionService] Result type: {type(result)}")
+            
+            description = ""
             if isinstance(result, (list, tuple)) and len(result) > 0:
-                return str(result[0])
-            return str(result)
+                description = str(result[0])
+                logging.info(f"[VisionService] Description extracted (length: {len(description)}).")
+            else:
+                description = str(result)
+                logging.info(f"[VisionService] Result was not a list/tuple, using string representation.")
+            
+            logging.debug(f"[VisionService] Description snippet: {description[:200]}...")
+            return description
+
         except Exception as e:
-            logging.error(f"Vision API call failed: {e}")
+            logging.error(f"[VisionService] Vision API call failed for {image_path}: {e}")
+            logging.error(traceback.format_exc())
             return f"Error describing screenshot: {str(e)}"
 
 vision_service = VisionService()