[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
619
stars
76
forks
source link
Performance difference between table 7 and table 11 in the paper. #11
Hi, is there any different configuration in tabel 7 and tabel 11 of the paper? In table 7, the mAP/NDS of TransFusion is 65.6/69.7.
In table 11, the mAP/NDS of TransFusion is 67.5/71.3. Thanks~
Thanks for your interest in our work. The results in Table 7 and Table 11 are achieved using different training schedules:
Table 7: train the first stage for 12 epochs without the fade strategy to get TransFusion-L, then train the second stage for 6 epochs to get TransFusion. We use this setting for fast iteration
Table 11: train the first stage for 20 epochs with fade strategy.
Hi, is there any different configuration in tabel 7 and tabel 11 of the paper? In table 7, the mAP/NDS of TransFusion is 65.6/69.7. In table 11, the mAP/NDS of TransFusion is 67.5/71.3. Thanks~