search

zhanggang001 / HEDNet

HEDNet (NeurIPS 2023) & SAFDNet (CVPR 2024 Oral)
Apache License 2.0
96 stars 9 forks source link

About the performance on the nuscenes validation set #22

Open zheng0819 opened 2 weeks ago

zheng0819 commented 2 weeks ago

Hello, I'm very interested in this safdnet work, when I reproduce it (8 RTX3090 GPUs), it performs 66.15% mAP and 70.41% NDS on nuscenes validation set,the performance is lower than the claimed 66.3% mAP and 71.0% NDS (especially NDS). I dont know the reason for this, is it because Transfusion_L head in OpenPCDet framework does not perform as well as mmdetection3d? Can you offer some advice on reproducing the work, much appreciated!

rubbish001 commented 1 week ago

他这个66.7%mAP，很难超越的，我的test点数比他高1.5%mAP，val都没有超过他

zheng0819 commented 1 week ago

他这个66.7%mAP，很难超越的，我的test点数比他高1.5%mAP，val都没有超过他

您好，请问您复现的NDS也低的很明显吗

rubbish001 commented 1 week ago

我就一刷点的，他的我没有跑哈，我的Co-Fix3D效果比他的好一点哈，希望作者遇到了给个7+分啊，如果这篇能中的话，我就开源，我是4卡跑的，让你们8卡的帮我跑一波，看看val到底能刷到多少。而且我现在的融合的一篇就可以弄出去了，我现在融合的val刷到了72.0%mAP，73.7%NDS。

rubbish001 commented 4 days ago

看看我的工程，吊打目前所有的3D目标检测 哈哈 https://github.com/rubbish001/Co-Fix3d