PJLab-ADG / LoGoNet

[CVPR2023] LoGoNet: Towards Accurate 3D Object Detection with Local-to-Global Cross-Modal Fusion
246 stars 16 forks source link

When will you release the code, and how long did you train the model? #1

Closed lmomoy closed 1 year ago

sankin97 commented 1 year ago

We are rushing into a new work, maybe in the next month. As for the training time for Waymo, LoGoNet-1f needs about 30 hours with 8 A100 GPUs and 64 batch sizes and LoGoNet-5f needs about 64 hours with 8 A100 GPUs and 64 batch sizes. While about 10 hours with 4 A100 GPUs, 8 batch sizes and 80epoch for training LoGoNet on KITTI.