search

dvlab-research / UVTR

Unifying Voxel-based Representation with Transformer for 3D Object Detection (NeurIPS 2022)
227 stars 17 forks source link

nuscene preprocess #19

Closed GondorFu closed 1 year ago

GondorFu commented 1 year ago

It takes a long time on my device for preprocess nuscene dataset. Can you make the preprocessing results available for direct download?

Thank you so much!

yanwei-li commented 1 year ago

Hi, I share the preprocessed file here. Maybe you can try it. But please make sure the data path is set to ./data/nuscenes.

GondorFu commented 1 year ago

Hi, I share the preprocessed file here. Maybe you can try it. But please make sure the data path is set to ./data/nuscenes.

According to your preprocessing code, after generate the pkl files, the code run the create_groundtruth_database function. Actually, this is the most time-consuming part. It seems generate the data/nuscenes/nuscenes_unified_gt_database floder, wouldn't these be used in training?

yanwei-li commented 1 year ago

This is used for modality fusion and lidar-based settings. Please ignore it if you already have generated GT database before or you only want to test on camera-based settings.