Danila Rukhovich

I don't have a good answer here. We also noticed that random initialization is not good enough for both instance and semantic segmentation. Not sure if there is anything particularly...

are you solving instance or semantic segmentation? for semantic segmentation you simply should not calculate loss for unlabeled points. for instance segmentation unlabeled points are also not a problem, as...

Yes I think if your valid classes are 0, ... n_classes - 1, than given a [n_classes](https://github.com/filaPro/oneformer3d/blob/main/oneformer3d/transforms_3d.py#L123) label for a superpoint will result in ignoring it in instance segmentation.

Looks like you are training with very low batch size. In this case if all scenes in the batch don't have ground truth objects, there may be NaN in the...

Hi @clawCa , Unfortunately we don't support visualization in this repo, please see #57.

We have S different predicted instance mask. Also we have S * n_classes scores for them. Then we select topk best scores from these S * n_classes. Then we select...

> but why the same mask can correspond to several scores and different labels As I remember it just has a minor positive impact on mAP metric (may be around...

Can you try with python package versions closer to the ones recommended in our Dockerfile? Including pytorch 1.11 -> 1.13.

Unfortunately, removing this line is not the solutions, as we need all imports from './oneformer3d' directory. I can only recommend to use Docker or virtualenv with exactly the same version...

We never tried multi-gpu for oneforner3d, so probably yes, multiplying should be fine, but i'm not sure if after it the metrics will be reproduced exactly.