inpainting task with salientbboxtrack mode not working as expected

[junhyeokahn]

Describe the issue

Running the preprocessing script with python vace/vace_preproccess.py --task inpainting --mode salientbboxtrack --video assets/videos/test.mp4 yields a contour mask (cat silhouette) rather than a tracking bounding-box mask. The output looks similar to salientmasktrack:

Expected behavior

For --mode salientbboxtrack, I expect the subject (cats) to be masked using a tracked bounding box over time (rectangular region), suitable for inpainting within that box.

Thanks for maintaining this excellent library. Happy to test a patch or provide more logs if helpful.