Person Translation Deduction

JiangWenPL / multiperson

Code repository for the paper: "Coherent Reconstruction of Multiple Humans from a Single Image" in CVPR'20

https://jiangwenpl.github.io/multiperson/

376 stars 50 forks source link

Person Translation Deduction #22

Closed FishWoWater closed 2 years ago

FishWoWater commented 3 years ago

https://github.com/JiangWenPL/multiperson/blob/4d3dbae945e22bb1e270521b061a837976699685/mmdetection/mmdet/models/losses/smpl_loss.py#L171 to https://github.com/JiangWenPL/multiperson/blob/4d3dbae945e22bb1e270521b061a837976699685/mmdetection/mmdet/models/losses/smpl_loss.py#L176

Could you give a brief explanation about L171 to L176? It's an approximation that depth d = 2 focal_length / (s max(box_w, box_h)), right? Because generally small person means larger depth.

JiangWenPL commented 2 years ago

Yes, in fact it was discussed in the supplementary.