latest changes

app.py

@@ -41,7 +41,8 @@ def sample_frames(frames_list, target_count):
     if len(frames_list) <= target_count:
         return frames_list
     indices = np.linspace(0, len(frames_list) - 1, target_count, dtype=int)
-    sampled = [frames_list(int(i)) for i in indices] # Corrected sampling
+    # FIX: Corrected list indexing from () to []
+    sampled = [frames_list[int(i)] for i in indices]
     return sampled
 
 def live_predict_stream(image_np_array):
@@ -74,7 +75,8 @@ def live_predict_stream(image_np_array):
                 if len(frames_for_model) < MODEL_INPUT_NUM_FRAMES:
                     prediction_result = "Error: Not enough frames for model."
                     status_message = "Error during frame sampling."
-                    app_state = "recording"
+                    print(f"ERROR: Insufficient frames for model input: {len(frames_for_model)}/{MODEL_INPUT_NUM_FRAMES}")
+                    app_state = "recording" # Reset to recording state
                     raw_frames_buffer.clear()
                     current_clip_start_time = time.time()
                     last_prediction_completion_time = time.time()
@@ -88,7 +90,7 @@ def live_predict_stream(image_np_array):
                     logits = outputs.logits
 
                 predicted_class_id = logits.argmax(-1).item()
-                predicted_label = model.config.id2label.get(predicted_class_id, "Unknown") # Handle potential missing label
+                predicted_label = model.config.id2label.get(predicted_class_id, "Unknown")
                 confidence = torch.nn.functional.softmax(logits, dim=-1)[0][predicted_class_id].item()
 
                 prediction_result = f"Predicted: {predicted_label} (Confidence: {confidence:.2f})"
@@ -175,12 +177,20 @@ with gr.Blocks() as demo:
             Use this API endpoint to send base64-encoded frames for prediction.
             """
         )
+        # Re-adding a slightly more representative API interface
+        # Gradio's automatic API documentation will use this to show inputs/outputs
         gr.Interface(
-            fn=lambda x: "API endpoint is active",
-            inputs=gr.Textbox(label="Input (Base64 JSON)"),
-            outputs=gr.Textbox(label="Status"),
-            title="API Status (Details in app.py)" # Minimal UI for API tab
+            fn=lambda frames_list: f"Received {len(frames_list)} frames. This is a dummy response. Integrate predict_from_frames_api here.",
+            inputs=gr.Json(label="List of Base64-encoded image strings"),
+            outputs=gr.Textbox(label="API Response"),
+            live=False,
+            allow_flagging="never" # For API endpoints, flagging is usually not desired
         )
+        # Note: The actual `predict_from_frames_api` function is defined above,
+        # but for a clean API tab, we can use a dummy interface here that Gradio will
+        # use to generate the interactive API documentation. The actual API call
+        # from your local script directly targets the /run/predict_from_frames_api endpoint.
+
 
 if __name__ == "__main__":
     demo.launch()