search

dvlab-research / UVTR

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

image gtaug difference with pointaugmenting #24

Open AndyYuan96 opened 1 year ago

AndyYuan96 commented 1 year ago

Hi,I see in paper that you say: 'In this work, we follow previous studies [50, 7] and use a unified approach for GT-sampling', but after I reading the point-augmenting code and your code, I still find some difference between your unified approach and point-augmenting approach, for example, when generate image gtaug, you only keep one image gtaug for one 3d box, but point-augmenting keep 6 image gtaug in nuscenes, can I ask did your approach is better than Pointaugmenting.

yanwei-li commented 1 year ago

Hi, sorry for the late reply. Could you please tell me the detailed code line? So, I can check it.