dvlab-research / FocalsConv

Focal Sparse Convolutional Networks for 3D Object Detection (CVPR 2022, Oral)
https://arxiv.org/abs/2204.12463
Apache License 2.0
371 stars 35 forks source link

Does FocalsConv supports spconv1.2.1? #25

Closed Raiden-cn closed 2 years ago

Raiden-cn commented 2 years ago

The released ckpt file was trained on OpenPCDet 0.3.0, but it applied spconv2.0, right? When I use latest pcdet(0.6.0), I encounter this error (using spconv 1.2.1) as following: AttributeError: 'SparseConvTensor' object has no attribute 'replace_feature' I don't know whether this pcdet 0.5.2 supports spconv 1.2.1. Anyway, how to run FocalsConv on spconv 1.2.1 is what I am concerned.

yukang2017 commented 2 years ago

Our implementation is previously based on spconv 1x. That is why the checkpoints are based on spconv 1x.

You can follow this doc to check the difference between the usage of spconv 1x and spconv 2x. And modify the code to spconv 1x. This is not a hard job.

Raiden-cn commented 2 years ago

get it now. Thanks!