Update app.py

app.py

@@ -371,6 +371,10 @@ def track(
 
     else:
         queries = query_points_tensor
+        xy = get_points_on_a_grid(15, video_input.shape[3:], device=device)
+        queries__ = torch.cat([torch.zeros_like(xy[:, :, :1]), xy], dim=2).to(device)  #
+        num_tracks = queries.shape[1]
+        queries = torch.cat([queries,queries__],dim=1)
         add_support_grid=True
 
     model(video_chunk=video_input, is_first_step=True, grid_size=0, queries=queries, add_support_grid=add_support_grid)
@@ -393,8 +397,11 @@ def track(
 
     pred_tracks = (pred_tracks * torch.tensor([video_preview.shape[2], video_preview.shape[1]]).to(device) / torch.tensor([VIDEO_INPUT_RESO[1], VIDEO_INPUT_RESO[0]]).to(device))
     
-    vis = Visualizer(save_dir="./saved_videos", pad_value=120, linewidth=3)
-    painted_video = vis.visualize(torch.tensor(video_preview).permute(0, 3, 1, 2)[None].to(pred_tracks.device), pred_tracks, pred_visibility, save_video=False)[0].permute(0, 2, 3, 1).cpu().numpy()
+    vis = Visualizer(save_dir="./saved_videos", pad_value=0, linewidth=1, tracks_leave_trace=-1)
+    segm_mask = torch.zeros(queries.shape[1]).to(device)
+    segm_mask[:num_tracks] = 1
+    print('segm_mask',segm_mask.shape, segm_mask)
+    painted_video = vis.visualize(torch.tensor(video_preview).permute(0, 3, 1, 2)[None].to(pred_tracks.device), pred_tracks, pred_visibility, save_video=False, segm_mask=segm_mask, compensate_for_camera_motion=True)[0].permute(0, 2, 3, 1).cpu().numpy()
     
     # painted_video = paint_point_track(video_preview,tracks,pred_occ,colors)