pretrained model for nuscenes - missing layer

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

733 stars 64 forks source link

pretrained model for nuscenes - missing layer #64

Closed YMarrakchi closed 3 months ago

YMarrakchi commented 3 months ago

first of all, thanks a lot for sharing the code and the weighs for this nice work.

I have just noticed that the provided checkpoint for nuscenes contained 541 keys while the network has 542 entries. The missing layer is 'backbone_3d.conv_input.0.weight' which is the very first layer of the model.

Could you please fix the checkpoint and reupload it?