waymo-open-dataset
waymo-open-dataset copied to clipboard
gt.bin for validation in ver 1.3
Thanks for the great competition and Dataset! I'm trying to use Waymo Perception Dataset 1.3.2 in MMDetection3D, but I'm still wondering if gt.bin for validation in ver 1.3 corresponds to laser_gt_objects.bin here. If not, where can I find it? Thanks!
Happened to see this issue :)
What we used before is this gt.bin version in Waymo 1.2.0. I observe that your mentioned new version of gt.bin is a little larger than the previous one, possibly because of the introduction of new contents like seg labels or something else. So I guess you can have a try with that. Looking forward to your feedback.
BTW, I guess the "laser" in the file name mainly distinguishes it from camera-only cases, where there is only a part of objects that can be detected in the camera views. We will release our produced gt labels for this task afterward.
I met this problem, too. In fact, laser_gt_objects.bin in ver1.3 can not evaluate vision-only method properly. I'm wondering what will be the rules to create camera-only gt.bin out of the val set, in your opinion @Tai-Wang .
I met this problem, too. In fact, laser_gt_objects.bin in ver1.3 can not evaluate vision-only method properly. I'm wondering what will be the rules to create camera-only gt.bin out of the val set, in your opinion @Tai-Wang .
Did you completed it on your project detr3d-zlt. I saw the filter program - filter_cam_only_gt_bin.py on the project. And do you think laser_gt_objects.bin
corresponds to gt.bin
?
I met this problem, too. In fact, laser_gt_objects.bin in ver1.3 can not evaluate vision-only method properly. I'm wondering what will be the rules to create camera-only gt.bin out of the val set, in your opinion @Tai-Wang .
Did you completed it on your project detr3d-zlt. I saw the filter program - filter_cam_only_gt_bin.py on the project. And do you think
laser_gt_objects.bin
corresponds togt.bin
?
Hi Yuanyuan,
I have some experimental code here: https://github.com/open-mmlab/mmdetection3d/pull/2179
Hi @ZLTJohn,
I am delighted to have received your response, and I appreciate your contribution to the open-sources community. The code you've provided has been very helpful to me. I sincerely hope that it can be merged into MMDectection3D one the pull request is finalized.