Errors occur in training pipeline using my own dataset

[erty5852]

Hello, I got some problems when I tried to train models with my own dataset. I followed tutorial 9 in docs, and dataset is ready in correct formats. After that I modified config in ./configs/base/datasets/nus-3d.py (the models configs I used are based on nuscenes) as tutorial mentions:

dataset_A_train = dict( type='Custom3DDataset', ann_file = 'annotation.pkl', pipeline=train_pipeline )

After I ran the tools/train.py ..., it reported bugs like:

File "/home/user/mmdetection3d/mmdet3d/datasets/pipelines/loading.py", line 202, in call ts = results['timestamp'] KeyError: 'timestamp'

My first thinking is I need to modify type='LoadPointsFromMultiSweeps' in train_pipeline in configs, because line 202 comes out from class LoadPointsFromMultiSweeps. Besides I don't have "sweeps" as nuscenes in my own dataset for sure, so it's perhaps meaningless for this part in the pipeline in my case.

But after I removed the dict of 'LoadPointsFromMultiSweeps' or make the dict empty (sorry i just give it a try), another error issues come out like RuntimeError: mat1 dim 1 must match mat2 dim 0. Bugs are from torch, so I really have no ideas to deal with it.

So I would like to ask if somebody knows how to handle it in my case, if this errors comes out because of LoadPointsFromMultiSweeps truly, and how to fix this so that the training pipeline succeeds. Thanks a lot!!!

[ZCMax]

RuntimeError: mat1 dim 1 must match mat2 dim 0. means that the rotation matrix dimension does not match, maybe like a situation: [3x3] x [4x3]. Besides, can you provide the error log file? Or it would be better if you can provide your annotation file if it's convienient.