Closed kylelee82 closed 4 years ago
@kylelee82 what is the state of the PR? Is it ready to be merged?
@ternaus the changes are complete and should fix the issue indicated (test was done by running on the dataset). If there are any concerns with removing the assert
I'd be happy to get that included somehow.
BACKGROUND
This PR fixes the KITTI conversion issue related to https://github.com/lyft/nuscenes-devkit/issues/73 for camera and LIDAR ego pose differences in the Lyft dataset, which directly affect 2D image bounding boxes in the post-converted KITTI labels.
Example (7dbbd3cba5220b616cd74d1ca0c4ad0b43ae4feb08dab84dcbbc5d7dfe9bb000):
[Without ego pose correction]![Figure_1](https://user-images.githubusercontent.com/7127211/69024979-eec80480-0979-11ea-9286-a26eaf8cc08b.png)
[With ego pose correction]![Figure_1 fixed](https://user-images.githubusercontent.com/7127211/69024985-f1c2f500-0979-11ea-9647-9d06a27676e5.png)
Testing:
export_kitti.py
was run on the Lyft dataset and visually checked for box alignment.