Mask Annotation

[Jankin Tang]

Do the mask annotations in the 144 test videos are obtained in the same way with the 4 mot-test videos? The way to generate the annotation, not about how many frames are annotated.

[luka.c...@gmail.com]

I am not sure if I understand what the question is about. So, the annotations in the VOTS2023 dataset are also sparsely annotated (not every frame is annotated), but the spacing between annotations may be different.

[Jankin Tang]

My question is how do you get the mask annotation for the 144 videos, not about the gap between the annotated frames.
For example, by manually annotation, or by SAM.

Besides, does the way for annotation in the 4 mot-test videos the same with the 144 videos?

[luka.c...@gmail.com]

I think the annotation approach is similar (I am not the one involved in annotation though). The process is a mix of automatic annotation, manual refinement and quality control. If you want to know more, I will prompt people who are doing it to answer here.

[Jankin Tang]

Thank you very much.

But i think i want to know more, so please prompt the person involved in this process.

[Ondřej Drbohlav]

The masks were manually annotated by professional annotators and checked by their supervisor.

The annotation tools involved the combinations of manual or automatic detection of bounding boxes, algorithmic segmentation (SAM, AlphaRefine, others) and, most of all, manual annotation.

Dne čtvrtek 25. května 2023 v 14:16:59 UTC+2 uživatel 175288...@gmail.com napsal: