Maksim Kolodiazhnyi

Hi, @meyerjo. It looks a little bit strange. But there are several recommendations below, that might help you to avoid this problem: 1) check that there are no other processes...

Please follow the instruction provided [here](https://github.com/filaPro/oneformer3d/tree/main/data/scannet#:~:text=The%20overall%20process%20for%20ScanNet%20could%20be%20achieved%20through%20the%20following%20script) and run `python batch_load_scannet_data.py` before.

Hi! Could you test the released model trained on 6 datasets only on s3dis? You only need to change this part of the [config](https://github.com/filaPro/unidet3d/blob/master/configs/unidet3d_1xb8_scannet_s3dis_multiscan_3rscan_scannetpp_arkitscenes.py): ``` val_dataloader = dict( batch_size=1, num_workers=1,...

Please download the preprocessed dataset from [here](https://huggingface.co/datasets/maksimko123/UniDet3D/blob/main/s3dis.tar.gz) and run inference again.

Hey! I believe you should use a config like this. It computes metrics on ScanNet across all 18 classes, but you can just remove any nonexistent classes when calculating the...