The performance of transfusion_nusc_voxel_LC

XuyangBai / TransFusion

[PyTorch] Official implementation of CVPR2022 paper "TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection with Transformers". https://arxiv.org/abs/2203.11496

Apache License 2.0

640 stars 77 forks source link

The performance of transfusion_nusc_voxel_LC #105

Closed Wang-pengfei closed 7 months ago

Wang-pengfei commented 8 months ago

Hi, Xuyang.

I'm currently working on reproducing TransFusion. After training for 20 epochs on LiDAR, I obtained results with mAP: 0.6463 and NDS: 0.6979, which are close to the original TransFusion-L.

However, when I continued training for 6 epochs for fusion, the first epoch yielded very low results with mAP: 0.3156 and NDS: 0.2860. Is this normal?

Shiming94 commented 2 months ago

Hi, @Wang-pengfei , did you solve this problem? I trained Transfusion_L with the original script, but my performance can only reach NDS around 67 and mAP around 60. Do you have any ideas on how I can improve this, or are there any other tricks I missed?