Use DETR3D to train my own dataset

[zhangbaoj]

Hello, I want to use the DETR3D algorithm to train on my own dataset. But now I can't label my dataset in the nusences dataset format, only in the kitii format. If I want to use the DETR3D algorithm to train the KITTI dataset, what should I modify. I hope you can give me a hint. Thank you very much.

[JoshuaScheuplein]

Hi, I would also be interested in how we can train DETR3D with a customized dataset. In the MMDetecion3D documentation, I read that it is possible to convert a customized dataset to the standard format. However, I don't know how to use this standard format in create_data.py

The doc proposes the following:

Once you prepared the raw data following our instruction, you can directly use the following command to generate training/validation information files.

python tools/create_data.py custom --root-path ./data/custom --out-dir ./data/custom --extra-tag custom

However, there is no option for 'custom' in the corresponding file create_data.py, so could you please clarify this approach a little bit? Thanks in advance!

[reynerliu]

@zhangbaoj have you solved it? my custom dataset is also KITTI format. And I changed the train config&dataset config to make sure the project can correctly run. And I got the bad result <grad_norm=nan,loss=0>. If you solved this problem,can you give me some advice about it?