more updates for preferences

app.py

@@ -349,6 +349,9 @@ def process_single_video(
         target_progress = np.linspace(0.0, 1.0, num=num_frames).tolist()
         success_label = [1.0 if prog > 0.5 else 0.0 for prog in target_progress]
 
+        # predict_last_frame_mask: server expects a list (1.0 per frame); omit and server pad_list_to_max can get None
+        predict_last_frame_mask = [1.0] * num_frames
+
         # Create Trajectory
         trajectory = Trajectory(
             task=task_text,
@@ -356,6 +359,7 @@ def process_single_video(
             frames_shape=frames_shape,
             target_progress=target_progress,
             success_label=success_label,
+            predict_last_frame_mask=predict_last_frame_mask,
             metadata={"source": "gradio_app"},
         )
 
@@ -468,6 +472,10 @@ def process_two_videos(
         success_label_a = [1.0 if prog > 0.5 else 0.0 for prog in target_progress_a]
         success_label_b = [1.0 if prog > 0.5 else 0.0 for prog in target_progress_b]
 
+        # predict_last_frame_mask: server expects a list (1.0 per frame); None causes pad_list_to_max to fail
+        mask_a = [1.0] * num_frames_a
+        mask_b = [1.0] * num_frames_b
+
         # Create trajectories
         trajectory_a = Trajectory(
             task=task_text,
@@ -475,6 +483,7 @@ def process_two_videos(
             frames_shape=frames_shape_a,
             target_progress=target_progress_a,
             success_label=success_label_a,
+            predict_last_frame_mask=mask_a,
             metadata={"source": "gradio_app", "trajectory": "A"},
         )
 
@@ -484,6 +493,7 @@ def process_two_videos(
             frames_shape=frames_shape_b,
             target_progress=target_progress_b,
             success_label=success_label_b,
+            predict_last_frame_mask=mask_b,
             metadata={"source": "gradio_app", "trajectory": "B"},
         )
 
@@ -724,7 +734,7 @@ with demo:
                         value=False,
                         info="If enabled, predict progress per frame rather than feeding the entire video at once",
                     )
-                    analyze_single_btn = gr.Button("Analyze Video", variant="primary")
+                    analyze_single_btn = gr.Button("Compute Progress", variant="primary")
 
                     gr.Markdown("---")
                     gr.Markdown("**OR Select from Dataset**")
@@ -977,72 +987,73 @@ with demo:
             )
 
         with gr.Tab("Preference Analysis"):
+            # Full-width row: two videos side by side
             with gr.Row():
-                with gr.Column():
-                    with gr.Row():
-                        video_a_input = gr.Video(label="Video A", height=250)
-                        video_b_input = gr.Video(label="Video B", height=250)
-                    task_text_dual = gr.Textbox(
-                        label="Task Description",
-                        placeholder="Describe the task",
-                        value="Complete the task",
-                    )
-                    analyze_dual_btn = gr.Button("Compare Videos", variant="primary")
-
-                    gr.Markdown("---")
-                    gr.Markdown("**OR Select from Dataset**")
-                    gr.Markdown("---")
-
-                    with gr.Accordion("📁 Video A - Select from Dataset", open=False):
-                        dataset_name_a = gr.Dropdown(
-                            choices=PREDEFINED_DATASETS,
-                            value="jesbu1/oxe_rfm",
-                            label="Dataset Name",
-                            allow_custom_value=True,
-                        )
-                        config_name_a = gr.Dropdown(
-                            choices=[], value="", label="Configuration Name", allow_custom_value=True
-                        )
-                        with gr.Row():
-                            refresh_configs_btn_a = gr.Button("🔄 Refresh Configs", variant="secondary", size="sm")
-                            load_dataset_btn_a = gr.Button("Load Dataset", variant="secondary", size="sm")
+                video_a_input = gr.Video(label="Video A", height=320)
+                video_b_input = gr.Video(label="Video B", height=320)
 
-                        dataset_status_a = gr.Markdown("", visible=False)
-                        with gr.Row():
-                            prev_traj_btn_a = gr.Button("⬅️ Prev", variant="secondary", size="sm")
-                            trajectory_slider_a = gr.Slider(
-                                minimum=0, maximum=0, step=1, value=0, label="Trajectory Index", interactive=True
-                            )
-                            next_traj_btn_a = gr.Button("Next ➡️", variant="secondary", size="sm")
-                        trajectory_metadata_a = gr.Markdown("", visible=False)
-                        use_dataset_video_btn_a = gr.Button("Use Selected Video for A", variant="secondary")
-
-                    with gr.Accordion("📁 Video B - Select from Dataset", open=False):
-                        dataset_name_b = gr.Dropdown(
-                            choices=PREDEFINED_DATASETS,
-                            value="jesbu1/oxe_rfm",
-                            label="Dataset Name",
-                            allow_custom_value=True,
-                        )
-                        config_name_b = gr.Dropdown(
-                            choices=[], value="", label="Configuration Name", allow_custom_value=True
-                        )
-                        with gr.Row():
-                            refresh_configs_btn_b = gr.Button("🔄 Refresh Configs", variant="secondary", size="sm")
-                            load_dataset_btn_b = gr.Button("Load Dataset", variant="secondary", size="sm")
-
-                        dataset_status_b = gr.Markdown("", visible=False)
-                        with gr.Row():
-                            prev_traj_btn_b = gr.Button("⬅️ Prev", variant="secondary", size="sm")
-                            trajectory_slider_b = gr.Slider(
-                                minimum=0, maximum=0, step=1, value=0, label="Trajectory Index", interactive=True
-                            )
-                            next_traj_btn_b = gr.Button("Next ➡️", variant="secondary", size="sm")
-                        trajectory_metadata_b = gr.Markdown("", visible=False)
-                        use_dataset_video_btn_b = gr.Button("Use Selected Video for B", variant="secondary")
+            task_text_dual = gr.Textbox(
+                label="Task Description",
+                placeholder="Describe the task",
+                value="Complete the task",
+            )
+            analyze_dual_btn = gr.Button("Compute Preference", variant="primary")
+
+            gr.Markdown("---")
+            gr.Markdown("**OR Select from Dataset**")
+            gr.Markdown("---")
+
+            with gr.Accordion("📁 Video A - Select from Dataset", open=False):
+                dataset_name_a = gr.Dropdown(
+                    choices=PREDEFINED_DATASETS,
+                    value="jesbu1/oxe_rfm",
+                    label="Dataset Name",
+                    allow_custom_value=True,
+                )
+                config_name_a = gr.Dropdown(
+                    choices=[], value="", label="Configuration Name", allow_custom_value=True
+                )
+                with gr.Row():
+                    refresh_configs_btn_a = gr.Button("🔄 Refresh Configs", variant="secondary", size="sm")
+                    load_dataset_btn_a = gr.Button("Load Dataset", variant="secondary", size="sm")
+
+                dataset_status_a = gr.Markdown("", visible=False)
+                with gr.Row():
+                    prev_traj_btn_a = gr.Button("⬅️ Prev", variant="secondary", size="sm")
+                    trajectory_slider_a = gr.Slider(
+                        minimum=0, maximum=0, step=1, value=0, label="Trajectory Index", interactive=True
+                    )
+                    next_traj_btn_a = gr.Button("Next ➡️", variant="secondary", size="sm")
+                trajectory_metadata_a = gr.Markdown("", visible=False)
+                use_dataset_video_btn_a = gr.Button("Use Selected Video for A", variant="secondary")
+
+            with gr.Accordion("📁 Video B - Select from Dataset", open=False):
+                dataset_name_b = gr.Dropdown(
+                    choices=PREDEFINED_DATASETS,
+                    value="jesbu1/oxe_rfm",
+                    label="Dataset Name",
+                    allow_custom_value=True,
+                )
+                config_name_b = gr.Dropdown(
+                    choices=[], value="", label="Configuration Name", allow_custom_value=True
+                )
+                with gr.Row():
+                    refresh_configs_btn_b = gr.Button("🔄 Refresh Configs", variant="secondary", size="sm")
+                    load_dataset_btn_b = gr.Button("Load Dataset", variant="secondary", size="sm")
+
+                dataset_status_b = gr.Markdown("", visible=False)
+                with gr.Row():
+                    prev_traj_btn_b = gr.Button("⬅️ Prev", variant="secondary", size="sm")
+                    trajectory_slider_b = gr.Slider(
+                        minimum=0, maximum=0, step=1, value=0, label="Trajectory Index", interactive=True
+                    )
+                    next_traj_btn_b = gr.Button("Next ➡️", variant="secondary", size="sm")
+                trajectory_metadata_b = gr.Markdown("", visible=False)
+                use_dataset_video_btn_b = gr.Button("Use Selected Video for B", variant="secondary")
 
-                with gr.Column():
-                    result_text = gr.Markdown("")
+            gr.Markdown("---")
+            gr.Markdown("### Preference result")
+            result_text = gr.Markdown("")
 
             # State variables for datasets
             current_dataset_a = gr.State(None)