edit app to use frame steps

app.py

@@ -319,6 +319,7 @@ def process_single_video(
     task_text: str = "Complete the task",
     server_url: str = "",
     fps: float = 1.0,
+    use_frame_steps: bool = False,
 ) -> Tuple[Optional[str], Optional[str]]:
     """Process single video for progress and success predictions using eval server."""
     # Get server URL from state if not provided
@@ -365,6 +366,8 @@ def process_single_video(
 
         # Build payload and send to server
         files, sample_data = build_payload([progress_sample])
+        # Add use_frame_steps flag to sample_data
+        sample_data["use_frame_steps"] = use_frame_steps
         response = post_batch_npy(server_url, files, sample_data, timeout_s=120.0)
 
         # Process response
@@ -426,6 +429,7 @@ def process_two_videos(
     prediction_type: str = "preference",
     server_url: str = "",
     fps: float = 1.0,
+    use_frame_steps: bool = False,
 ) -> Tuple[Optional[str], Optional[str], Optional[str]]:
     """Process two videos for preference, similarity, or progress prediction using eval server."""
     # Get server URL from state if not provided
@@ -523,6 +527,8 @@ def process_two_videos(
 
             # Build payload and send to server
             files, sample_data = build_payload([progress_sample_a, progress_sample_b])
+            # Add use_frame_steps flag to sample_data
+            sample_data["use_frame_steps"] = use_frame_steps
             response = post_batch_npy(server_url, files, sample_data, timeout_s=120.0)
 
             # Process response
@@ -761,6 +767,11 @@ with demo:
                         step=0.1,
                         info="Frames per second to extract from video (higher = more frames)",
                     )
+                    use_frame_steps_single = gr.Checkbox(
+                        label="Use Frame Steps",
+                        value=False,
+                        info="Process frames incrementally (0:1, 0:2, 0:3, etc.) for autoregressive predictions",
+                    )
                     analyze_single_btn = gr.Button("Analyze Video", variant="primary")
 
                     gr.Markdown("---")
@@ -998,7 +1009,7 @@ with demo:
 
             analyze_single_btn.click(
                 fn=process_single_video,
-                inputs=[single_video_input, task_text_input, server_url_state, fps_input_single],
+                inputs=[single_video_input, task_text_input, server_url_state, fps_input_single, use_frame_steps_single],
                 outputs=[progress_plot, info_output],
                 api_name="process_single_video",
             )
@@ -1027,6 +1038,11 @@ with demo:
                         step=0.1,
                         info="Frames per second to extract from videos (higher = more frames)",
                     )
+                    use_frame_steps_dual = gr.Checkbox(
+                        label="Use Frame Steps (Progress mode only)",
+                        value=False,
+                        info="Process frames incrementally for progress predictions",
+                    )
                     analyze_dual_btn = gr.Button("Compare Videos", variant="primary")
 
                     gr.Markdown("---")
@@ -1476,7 +1492,7 @@ with demo:
 
             analyze_dual_btn.click(
                 fn=process_two_videos,
-                inputs=[video_a_input, video_b_input, task_text_dual, prediction_type, server_url_state, fps_input_dual],
+                inputs=[video_a_input, video_b_input, task_text_dual, prediction_type, server_url_state, fps_input_dual, use_frame_steps_dual],
                 outputs=[result_text, video_a_display, video_b_display],
                 api_name="process_two_videos",
             )