search

ADLab-AutoDrive / BEVHeight

An official code release of our CVPR'23 paper, BEVHeight
MIT License
200 stars 25 forks source link

During the evaluation, we primarily focus on the prediction & GT within [0, -51.2, -10, 135.0, 51.2, 10], please refer to the latest code base. #28

Closed mayechi closed 7 months ago

mayechi commented 1 year ago 
          During the evaluation, we primarily focus on the prediction & GT within [0, -51.2, -10, 135.0, 51.2, 10], please refer to the latest code base.

Originally posted by @yanglei18 in https://github.com/ADLab-AutoDrive/BEVHeight/issues/23#issuecomment-1683424869 I am doubt about it where I saw the post_center_range=[0.0, -61.2, -10.0, 122.4, 61.2, 10.0] and point_cloud_range=[0, -51.2, -5, 102.4, 51.2, 3] in every config of dair-v2x, so why you used [0, -51.2, -10, 135.0, 51.2, 10] during the evaluation?

yanglei18 commented 1 year ago

What do you think I should do?

mayechi commented 1 year ago

First choice, i think you should filter the GT with the point_cloud_range=[0, -51.2, -5, 102.4, 51.2, 3] liking nuscence metirc filtering the GT with point_cloud_range=[-54, -54, -5.0, 54, 54, 3.0]. Second choice, i think you should modify the point_cloud_range=[0, -51.2, -5, 102.4, 51.2, 3] to [0, -51.2, -10, 135.0, 51.2, 3]. Because if the obs are farther than 102.4m and closer than 135.0m, you threw them away directly now.