search

SxJyJay / MSMDFusion

[CVPR 2023] MSMDFusion: Fusing LiDAR and Camera at Multiple Scales with Multi-Depth Seeds for 3D Object Detection
Apache License 2.0
167 stars 10 forks source link

Questions on Virtual Points #30

Open russellyq opened 6 months ago

russellyq commented 6 months ago

Hi,

Thanks for your great work ! When I am reading your codes, I am confused about how you generate the virtual points by a specific seed, as described in your paper.

Looking forward to your answer.

Thanks.

SxJyJay commented 6 months ago

Do you mean that the method of selecting specific seed points in the 2D image plane for lifting into 3D voxel space? We randomly select some pixels as seeds per instance.

russellyq commented 6 months ago

Do you mean that the method of selecting specific seed points in the 2D image plane for lifting into 3D voxel space? We randomly select some pixels as seeds per instance.

Hi, thanks for your response.

Yes. But we did not see how you do this process in your codes.

SxJyJay commented 6 months ago

Because we generate these virtual points offline, and I provide the pre-generated virtual points through BaiduYun.

russellyq commented 6 months ago

Thanks for your answer! I still have some questions regarding this.

  1. how do you generate these points offline ? Did you generate these points according to MVP ?
  2. how did you read these points into the framework? Also, I see that you generate voxel_2d in (https://github.com/SxJyJay/MSMDFusion/blob/7b5b2741e693ba8007c95e3e8951e4e67fbc47ed/mmdet3d/models/detectors/MSMDFusion.py#L382) what is the relationship between the voxel2d here and the pre-generated virtual points ?

Thanks for your answer ! Also, since I may have some more questions, can we share our wechat via email ( qiaoyan@cuhk.edu.hk ) and we can have more corporation in future.