TypeError: The annotations loaded from annotation file should be a dict, but got <class 'list'>! The intention of why not directly making "dict .pkl"?

[perrydoremi]

What is the feature?

Currently I'm implementing my own dataset in mmdetection3d for point cloud segmentation task. I got this error when testing pointnet2.

TypeError: The annotations loaded from annotation file should be a dict, but got <class 'list'>! when testing pointnet2.

I checked all the issues and found that the current .pkl file should be a "dict" not a "list" as before, but I'm so confused how this could happen cause I just forked the newest mmdet3d and implemented my dataset from scratch. So I went through all the files that related to data preparation, and found something that's difficult to understand:

Take s3dis dataset for eg: In create_data.py, why in s3dis_data_prep function, a update_pkl_infos is compulsorily used to convert "list .pkl" to "dict .pkl" instead of directly making "dict .pkl" files in indoor.create_indoor_info_file ? Could you pls help me explain the intention here?

Really appreciate mmdet3d's dedicated work so far, and many thx!!!

Any other context?

No response

[ThaiNguyenMonash]

how to change that to dict?

[temex12]

how to change that to dict?

I believe you can use the update_infos_to_v2.py to change it. As below: python tools/dataset_converters/update_infos_to_v2.py --dataset ${DATA_SET} --pkl-path ${PKL_PATH} --out-dir ${OUT_DIR} Example usage: python update_infos_to_v2.py --dataset kitti --pkl-path \data\kitti\kitti_infos_train.pkl --out-dir \data\kitti_updated