search

Thinklab-SJTU / Bench2Drive

Closed-loop multi-ability evaluation of end-to-end autonomous driving algorithms
Apache License 2.0
1.09k stars 77 forks source link

Empty lanes #57

Open Timer-x opened 4 weeks ago

Timer-x commented 4 weeks ago

I find many frames with empty lane when i try to obtain map data from B2D_vad_dataset.py. I don't know if there is something wrong when i prepare the b2d dataset.

zhiyuanzzz commented 4 weeks ago

Please provide more details so we can find the problem.

Timer-x commented 3 weeks ago

image

this error occurred because there is no lane

Timer-x commented 3 weeks ago

image

I download map data from baidu cloud, and process all the data using (https://github.com/Thinklab-SJTU/Bench2DriveZoo/blob/a630480bdc0222382c41dd80115f690bbdb58505/mmcv/datasets/prepare_B2D.py#L400).

zhiyuanzzz commented 3 weeks ago

get_map_info in B2D_vad_dataset.py obtains lanes within a certain range of the ego vehicle, so please ensure that you are using the correct ego position and config. If you want to train VAD on Bench2Drive, it is recommended to use the VAD model in our repo ( https://github.com/Thinklab-SJTU/Bench2DriveZoo/tree/uniad/vad ) and follow the documents to prepare data and train model.

Timer-x commented 3 weeks ago

I use your official code. https://github.com/Thinklab-SJTU/Bench2DriveZoo/blob/a630480bdc0222382c41dd80115f690bbdb58505/mmcv/datasets/B2D_vad_dataset.py#L299 and add print(gt_bboxes.fixed_num_sampled_points) to check the data but failed. Can you double check it again. Thanks a lot.

ChipsICU commented 3 weeks ago

I use your official code. https://github.com/Thinklab-SJTU/Bench2DriveZoo/blob/a630480bdc0222382c41dd80115f690bbdb58505/mmcv/datasets/B2D_vad_dataset.py#L299 and add print(gt_bboxes.fixed_num_sampled_points) to check the data but failed. Can you double check it again. Thanks a lot.

@Timer-x maybe you can check this issue, https://github.com/Thinklab-SJTU/Bench2Drive/issues/62

Timer-x commented 3 weeks ago

I use your official code. https://github.com/Thinklab-SJTU/Bench2DriveZoo/blob/a630480bdc0222382c41dd80115f690bbdb58505/mmcv/datasets/B2D_vad_dataset.py#L299 and add print(gt_bboxes.fixed_num_sampled_points) to check the data but failed. Can you double check it again. Thanks a lot.

@Timer-x maybe you can check this issue, #62

Thank you, but I find this problem in open loop training. Have you attempt to visualize the map data when you train the vad?