Track-Anything
Track-Anything copied to clipboard
masks displayed from multiple frames
Hi,
Nice work!
- Using the demo app, whenever one clicks on a "Add mask" and then switches to a new frame, and then adds another mask, all masks from all time steps are displayed. Perhaps it would be better to display only masks associated to a particular frame?
- If multiple masks are selected of the same object over a longer video sequence, they interpreted as separate tracked objects and not merged. How can one select in the UI multiple occurrences of the same track?
Also, I often get the following error
in generate_video_from_frames
frames = torch.from_numpy(np.asarray(frames))
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (200,) + inhomogeneous part.