search

dvlab-research / VoxelNeXt

VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking (CVPR 2023)
https://arxiv.org/abs/2303.11301
Apache License 2.0
735 stars 64 forks source link

IoU related loss #41

Open DezeZhao opened 1 year ago

DezeZhao commented 1 year ago

Hi, yukang @yukang2017, yanwei @yanwei-li, I am confused about the iou related loss you used in dense head(IoULossSparse and IoURegLossSparse). I reproduce your model on my devices, but I find IoURegLossSparse does not seem to going down, it keeps almost the same value in the training process. I use IoURegLoss in another model, I trap into the same problem(not decrease). Could you please show me your intention to use the loss?

Additionally, why use IoULossSparse and IoURegLossSparse repsectively? I can't understand the purpose of their use. Looking forward to your reply, thank you.

Best regards. deze

Gorazor commented 1 year ago

Have the same question too! why use IoULossSparse and IoURegLossSparse repsectively