open-mmlab
KITTI visualization bug
python.exe C:/Users/Guo/PycharmProjects/mmdetection3d/tools/misc/browse_dataset.py configs/base/datasets/kitti-mono3d.py --task mono-det --aug --output-dir ./kitti_vis --online
image of 65/3712
It seems some boxes outside of the image are not filtered and the projection is from boxes behind the image plane. Do you locate the bug successfully? If not, we will reproduce and fix it ASAP.
It seems some boxes outside of the image are not filtered and the projection is from boxes behind the image plane. Do you locate the bug successfully? If not, we will reproduce and fix it ASAP.
I have not located that, but during locating I found another bug that the 2D bbox is not tightly attached to 3D bbox, the problem is in image_vis.py line 199, it use P2 to project with translation but the 3D bbox coordination has been translated to camera 2 coordinate system, so I simply add cam2img[:, -1] = 0 in line 200.
Thanks for your report. Actually, we find another tricky bug that may cause some boxes not to be correctly filtered here. We should input a custom argument for image size on KITTI for this function. We will fix it afterward, but it may need other regression validation. If convenient, you can also try to fix it and have a look at this issue again.
It seems some boxes outside of the image are not filtered and the projection is from boxes behind the image plane. Do you locate the bug successfully? If not, we will reproduce and fix it ASAP.
I have not located that, but during locating I found another bug that the 2D bbox is not tightly attached to 3D bbox, the problem is in image_vis.py line 199, it use P2 to project with translation but the 3D bbox coordination has been translated to camera 2 coordinate system, so I simply add cam2img[:, -1] = 0 in line 200.
Maybe the box3d is better to be unified into the camera0 system and the projection would be more smooth. This is really a subtle problem lying in the definition of coordinate systems on KITTI. We will also consider adjusting it afterward.
@ZCMax Please have a look at this issue. We may make a plan to adjust related codes and make a regression for validation.
