search

traveller59 / second.pytorch

SECOND for KITTI/NuScenes object detection
MIT License
1.72k stars 722 forks source link

kitti 3d car detection performance #37

Open teddybuy opened 5 years ago

teddybuy commented 5 years ago

How to tweak the hyper-parameters to get closer to the good performance listed in the kitti leaderboard? Thanks!

http://www.cvlibs.net/datasets/kitti/eval_object_detail.php?&result=9da5eb9e19aa203a936e2867b44cb24b92f972ac

traveller59 commented 5 years ago

What's your test performance? This project can directly use pretrained model (car) to generate same labels in that submit. But that model is submitted in 4.28 and the code is changed a lot today. The major change is sample functions and noise functions. may be you can tune these parameters.