pts_semantic_mask_path key is not exist in the nuscenes dataset for occupancy prediction task

[SeaBird-Go]

Hello, when I run the TPVFormer project by the following command:

bash tools/dist_test.sh projects/TPVFormer/configs/tpvformer_8xb1-2x_nus-seg.py checkpoints/tpvformer_pretrained_fcos3d_r101_dcn.pth 8

I encountered the error that:

  File "~/mmdetection3d/mmdet3d/datasets/transforms/loading.py", line 1060, in transform
    results = self._load_semantic_seg_3d(results)
  File "~/mmdetection3d/mmdet3d/datasets/transforms/loading.py", line 945, in _load_semantic_seg_3d
    pts_semantic_mask_path = results['pts_semantic_mask_path']
KeyError: 'pts_semantic_mask_path'

It seems the pts_semantic_mask_path key not exist in the nuscenes dataset. And even though I create this dataset by following the prepare dataset step, it still fails.

[sunjiahao1999]

It seems to be that your pkl file does not have pts_semantic_mask_path in it. You can refer to this guide to regenerate nuscenes_infos_train.pkl. You have to make sure to additionally download the nuScenes lidarseg.

[sunjiahao1999]

btw, the checkpoint you are using is for training TPVFormer, you should use this checkpoint to validate the metrics.