search

XuyangBai / TransFusion

[PyTorch] Official implementation of CVPR2022 paper "TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection with Transformers". https://arxiv.org/abs/2203.11496
Apache License 2.0
619 stars 76 forks source link

Bug in training with config "transfusion_nusc_pillar_LC.py" #5

Closed YunzeMan closed 2 years ago

YunzeMan commented 2 years ago

Hi,

Thanks for releaseing the codebase! I can successful train the TransFusion-L with your provided config.

However, I encountered an error when I tried to train the TransFusion Lidar-image modalities. The config I used was the provided "transfusion_nusc_pillar_LC.py". The error was:

ValueError: could not broadcast input array from shape (448,796,3) into shape (900,1600,3)

I used load_from = None. I think this error can be easily reproduced. Could you please fix this or shed some light on the correct way to do it?

Thanks a lot!

xxlbigbrother commented 2 years ago

Hi,

Thanks for releaseing the codebase! I can successful train the TransFusion-L with your provided config.

However, I encountered an error when I tried to train the TransFusion Lidar-image modalities. The config I used was the provided "transfusion_nusc_pillar_LC.py". The error was:

ValueError: could not broadcast input array from shape (448,796,3) into shape (900,1600,3)

I used load_from = None. I think this error can be easily reproduced. Could you please fix this or shed some light on the correct way to do it?

Thanks a lot!

hi I change the code and it works!

修改代码