florence edge case

app.py

@@ -83,7 +83,14 @@ def run_florence(image, text_input):
         task=task_prompt, 
         image_size=(image.width, image.height)
     )
-    return parsed_answer[task_prompt]['bboxes'][0]
+    
+    bboxes = parsed_answer[task_prompt]['bboxes']
+    if not bboxes:
+        print(f"No objects detected for prompt: '{text_input}'. Trying with a default bounding box.")
+        # Return a default bounding box covering the entire image
+        return [0, 0, image.width, image.height]
+    
+    return bboxes[0]
 
 def remove_directory_contents(directory):
     for root, dirs, files in os.walk(directory, topdown=False):
@@ -227,7 +234,7 @@ demo = gr.Interface(
     This demo uses [Florence-2](https://huggingface.co/microsoft/Florence-2-large), to enable text-prompted object detection for [SAM2](https://github.com/facebookresearch/segment-anything).
 
     1. Upload a short video (< 10 sec , you can fork this space on larger GPU for longer vids) 
-    2. Describe the object to segment.
+    2. Describe the object to segment (The object should be visible in the first frame).
     3. Get your segmented video.
     """
 )