-
Hello everyone!
I am trying to train PointPillars on DAIR-V2X dataset. I have complete the dair_dataset.py similar to kitti_dataset.py and I can train without error. But when finish training, I get a…
-
I am trying to evaluate my tracker results using the nuscenes 'evaluate.py' script. I am using the v1.0-trainval version but I am only running my tracker on the first scene (of the pointPillars-val.js…
-
### Describe the issue
The feature extracted here `f_center = features[:, :, :2]` without `.clone()` will cause the first 2 columns in original `features` be replaced by `f_center` in the subsequent …
-
When I upgraded the Pytorch and Torchvision to 1.10.0 and 0.11.0 and then trained the pointpillars (https://github.com/traveller59/second.pytorch/blob/3aba19c9688274f75ebb5e576f65cfe54773c021/second/c…
-
@tianweiy , Have you tried to train Pointpillars model with the dynamic voxelizer in your realtime-version ?
Thanks,
Youshaa
-
Hi,
I tried training Pointpillars with fp16 on Waymo dataset but got nan output almost every iteration.
After some troubleshooting, I accidentally found out the nan is caused by the inf value a…
-
Hi,
thanks for providing this awesome repository, great work!
I want to get a subset of points from the whole point cloud which are part of one specific predicted bounding box. I am using the n…
-
Hello,
I am using the default hv_pointpillars_fpn_nus.py configuration to train nuscenes. The default pillar value is [0.25, 0.25, 8].
I want to experiment reducing the pillar size, for instance t…
-
Why do my running results of pedestrian with pointpillars in mmdet3d are so high compare the ranking of Kitti datasets.
stidk updated
2 years ago
-
When I use analyze_logs.py
```
['work_dirs/hv_pointpillars_secfpn_6x8_160e_kitti-3d-car/20211007_134844.log.json']
-----Analyze train time of work_dirs/hv_pointpillars_secfpn_6x8_160e_kitti-3d-car…