separate tab for video pointing

app.py

@@ -454,8 +454,10 @@ with gr.Blocks() as demo:
         # ── LEFT COLUMN: Inputs ──
         with gr.Column():
             with gr.Tabs() as input_tabs:
-                with gr.TabItem("Video Pointing & Tracking", id="video_tab") as video_tab:
-                    video = gr.Video(label="Input Video", elem_id="input_video", height=MAX_VIDEO_HEIGHT)
+                with gr.TabItem("Video Tracking", id="video_tracking_tab") as video_tracking_tab:
+                    video_tracking = gr.Video(label="Input Video", elem_id="input_video", height=MAX_VIDEO_HEIGHT)
+                with gr.TabItem("Video Pointing", id="video_pointing_tab") as video_pointing_tab:
+                    video_pointing = gr.Video(label="Input Video", elem_id="input_video_pointing", height=MAX_VIDEO_HEIGHT)
                 with gr.TabItem("Image(s) Pointing", id="image_tab") as image_tab:
                     images_input = gr.Gallery(
                         label="Input Images", elem_id="input_image", type="filepath", height=MAX_IMAGE_SIZE,
@@ -472,7 +474,7 @@ with gr.Blocks() as demo:
             with gr.Row():
                 submit_button = gr.Button("Submit", variant="primary", scale=3)
                 clear_all_button = gr.ClearButton(
-                    components=[video, images_input, input_text], value="Clear All", scale=1,
+                    components=[video_tracking, video_pointing, images_input, input_text], value="Clear All", scale=1,
                 )
 
         # ── RIGHT COLUMN: Outputs ──
@@ -499,16 +501,26 @@ with gr.Blocks() as demo:
                 output_annotations_img = gr.Gallery(label="Annotated Images", height=MAX_IMAGE_SIZE)
 
     # ── Examples ──
-    with gr.Group(visible=True) as video_examples_group:
-        gr.Markdown("### Video Examples")
+    with gr.Group(visible=True) as video_tracking_examples_group:
+        gr.Markdown("### Video Tracking Examples")
         gr.Examples(
             examples=[
                 ["example-videos/penguins.mp4", "Track all the penguins."],
-                # ["example-videos/backflips.mp4", "Count the number of backflips performed in this clip."],
                 ["example-videos/arena_basketball.mp4", "Track the players in yellow uniform in 1 fps."],
             ],
-            inputs=[video, input_text],
-            label="Video Pointing & Tracking Examples",
+            inputs=[video_tracking, input_text],
+            label="Video Tracking Examples",
+        )
+
+    with gr.Group(visible=False) as video_pointing_examples_group:
+        gr.Markdown("### Video Pointing Examples")
+        gr.Examples(
+            examples=[
+                ["example-videos/sports.mp4", "Point to players in white/blue jersey"],
+                ["example-videos/travel.mp4", "Point to standalone spiky towers"],
+            ],
+            inputs=[video_pointing, input_text],
+            label="Video Pointing Examples",
         )
 
     with gr.Group(visible=False) as image_examples_group:
@@ -523,12 +535,26 @@ with gr.Blocks() as demo:
         )
 
     # ── Tab switching: toggle visibility + track active tab ──
-    active_tab = gr.State("video")
+    active_tab = gr.State("video_tracking")
 
-    def _select_video_tab():
+    def _select_video_tracking_tab():
         return (
-            "video",
-            gr.update(visible=True),   # video_examples_group
+            "video_tracking",
+            gr.update(value=10),       # max_fps
+            gr.update(visible=True),   # video_tracking_examples_group
+            gr.update(visible=False),  # video_pointing_examples_group
+            gr.update(visible=False),  # image_examples_group
+            gr.update(visible=True),   # video_params_row
+            gr.update(visible=True),   # video_output_tabs
+            gr.update(visible=False),  # image_output_group
+        )
+
+    def _select_video_pointing_tab():
+        return (
+            "video_pointing",
+            gr.update(value=2),        # max_fps
+            gr.update(visible=False),  # video_tracking_examples_group
+            gr.update(visible=True),   # video_pointing_examples_group
             gr.update(visible=False),  # image_examples_group
             gr.update(visible=True),   # video_params_row
             gr.update(visible=True),   # video_output_tabs
@@ -538,15 +564,22 @@ with gr.Blocks() as demo:
     def _select_image_tab():
         return (
             "image",
-            gr.update(visible=False),  # video_examples_group
+            gr.update(),               # max_fps unchanged
+            gr.update(visible=False),  # video_tracking_examples_group
+            gr.update(visible=False),  # video_pointing_examples_group
             gr.update(visible=True),   # image_examples_group
             gr.update(visible=False),  # video_params_row
             gr.update(visible=False),  # video_output_tabs
             gr.update(visible=True),   # image_output_group
         )
 
-    tab_outputs = [active_tab, video_examples_group, image_examples_group, video_params_row, video_output_tabs, image_output_group]
-    video_tab.select(fn=_select_video_tab, outputs=tab_outputs)
+    tab_outputs = [
+        active_tab, max_fps,
+        video_tracking_examples_group, video_pointing_examples_group, image_examples_group,
+        video_params_row, video_output_tabs, image_output_group,
+    ]
+    video_tracking_tab.select(fn=_select_video_tracking_tab, outputs=tab_outputs)
+    video_pointing_tab.select(fn=_select_video_pointing_tab, outputs=tab_outputs)
     image_tab.select(fn=_select_image_tab, outputs=tab_outputs)
 
     _WARNING_STYLE = (
@@ -566,12 +599,13 @@ with gr.Blocks() as demo:
             return gr.update(value=html, visible=True)
         return gr.update(value="", visible=False)
 
-    def dispatch_submit(tab, user_text, video_path, input_images,
-                        fsm, mf, mfps, max_tok):
+    def dispatch_submit(tab, user_text, video_tracking_path, video_pointing_path,
+                        input_images, fsm, mf, mfps, max_tok):
         if tab == "image":
             text_out, img_gallery, pts = process_images(user_text, input_images, max_tok)
             return text_out, pts, gr.update(value="", visible=False), None, [], img_gallery
         else:
+            video_path = video_tracking_path if tab == "video_tracking" else video_pointing_path
             text_out, ann_video, ann_frames, pts = process_video(
                 user_text, video_path, fsm, mf, mfps, max_tok,
             )
@@ -580,7 +614,7 @@ with gr.Blocks() as demo:
 
     submit_button.click(
         fn=dispatch_submit,
-        inputs=[active_tab, input_text, video, images_input,
+        inputs=[active_tab, input_text, video_tracking, video_pointing, images_input,
                 frame_sample_mode, max_frames, max_fps, max_tok_slider],
         outputs=[output_text, output_points, output_warning, output_video, output_annotations, output_annotations_img],
     )

example-videos/sports.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:74dac8c7df7403f7e349c2b4037b117c4c7520edb9501e295665216d02e1ea6e
+size 6238226

example-videos/travel.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53e9cf8094d252dba4abdcf922dca79f1b0c454cc8fa18c0005e62068ddfa8bc
+size 9435014