Top-K for OVIS

[timmeinhardt]

The following line of code indicated that you are operating on a top-k with k=10 inference:

MinVIS/mask2former_video/video_maskformer_model.py

Line 260 in 9aa1337

scores_per_image, topk_indices = scores.flatten(0, 1).topk(10, sorted=False)

But since OVIS contains sequences with more than 20 objects you are by design not able to detect these. Increasing k could easily improve your OVIS results. Do you think this is correct?