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

A bug in 'RandomFlip3D' #1

Closed wangbo-zhao closed 2 years ago

wangbo-zhao commented 2 years ago

I keep all environments as the same as you. mmdet==2.10.0, mmcv==1.2.4, and I adopt the config file named "transfusion_waymo_voxel_LC.py"

If we get true in "results['flip']", then "mmcv.imflip(results[key], direction=results['flip_direction'])" will generate a np<class 'numpy.ndarray'> instead of a <class 'list'>, image

This will result a bug in image

XuyangBai commented 2 years ago

Hi, sorry for the late reply. Yes, I forget to turn the multiple images into a list, I will fix this bug soon. You can use this solution first. https://github.com/XuyangBai/TransFusion/issues/5

XuyangBai commented 2 years ago

Solved in https://github.com/XuyangBai/TransFusion/commit/ebd8ba4716bfbf324efc5164f53f9e9b8778556c, sorry for the inconvience.