Pointcept / Pointcept

Pointcept: a codebase for point cloud perception research. Latest works: PTv3 (CVPR'24 Oral), PPT (CVPR'24), OA-CNNs (CVPR'24), MSC (CVPR'23)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

About ElasticDistortion

RayYoh opened this issue · comments

Hi Xiaoyang. Are there any resources that can help understand ElasticDistortion data aug?
I have searched other open-sourced codes like PointGroup, Oneformer3D. All these code first div voxel_size or time voxel_scale, and then do the data augmentation use the same parameters while Pointcept uses different parameters. I am not sure whether this is a strong augmentation for the data and then further influences the final performance.

Hi, I think Spatio-Temporal Segmentation by Chrischoy is the first introduced ElasticDistortion. And I think the order of ElasticDistortion and GridSampling doesn't matter.

Yeah, it's true the order doesn't matter. But I have not understood the distortion parameters, I mean granularity and magnitude. Which have different scales. [[6, 40], [20, 160]] vs [[0.2, 0.4], [0.8, 1.6]]. I'm not sure whether I ignore some operation before ElasticDistortion.

Oh, actually, I also don't fully understand the meaning of these parameters. I just transfer them from Chris Choy's repo.

Alright, thanks for your reply.