-
pointpillars_kitti and second_kitti set `dir_limit_offset` in Anchor3DHead as default value `-pi / 2` while other datasets set the value as `-0.78554`(-pi / 4). Since the coordinate system has been re…
-
In multi-frame models, I'm assuming you map all Lidar points to the same (egocentric) frame. Is this done during preprocessing (building of the dataset), or is it done at training time by some part of…
-
I'm getting much slower inference time (roughly 50% slower) inference time on a pillars + center head model versus a regular pillars + anchor 3d head model. Theoretically the center head should be fas…
-
Hi, I notice that you have updated the anchor_generator type to `AlignedAnchor3DRangeGenerator` for pointpillars_kitti-3class, while pp_kitti-car, second_kitti-3class and second_kitti-car still use th…
-
Hi, I want to use K-fold cross validation for training PointPillars on nuScenes dataset. May I know where should I change the code?
-
I try to do the experiments in your paper using OpenPCDet.
For the PP-Lidar experi…
-
Traceback (most recent call last):
File "/media/5.5T_Harddisk/LIDAR/PointPillars_Nuscens/second.pytorch/second/create_data.py", line 25, in
nuscenes_data_prep('/media/5.5T_Harddisk/LIDAR/nusc…
-
Can anyone use OpenPCDet to reproduce the official results of Pointpillars in Cyclist and Pedestrian? If possible, can you provide me with the configuration file, thank you very much!
-
I noticed that PointPillars and CaDNN config files contain a "VOXEL_SIZE" attribute.
Are they used somewhere in the code ? What are the differences with the attribute defined in datasets config file…
-
Hi,
I am running kitti pointpillars model. And it seems there is bug in features computation. And it is there in below forward function inside the file mmdet3d/models/voxel_encoders/pillar_encoder.…