Where can I find the code of silhouette fitting

[FishWoWater]

Thanks for your nice work! As you mentioned, the sihouette fitting code is not released due to license reason, then where can I find this code? Thanks a lot!

[classner]

Sorry, it is not published and you'd have to implement it yourself.

[FishWoWater]

Alright, thank you!

[foermozhu]

Alright, thank you!

Hi,do you find the sihouette fitting code,I'm also studying this part. Look forward to your reply，thanks！！ 如果可以的话我们可以交流一下吗？

[FishWoWater]

Alright, thank you!

Hi,do you find the sihouette fitting code,I'm also studying this part. Look forward to your reply，thanks！！ 如果可以的话我们可以交流一下吗？

Hi! There are some possible solutions:

• You may implement the silhouette fitting by yourself, and you can refer to chamfer loss in pytorch3d.
• You may refer to nips17, which also uses silhouette as a supervision signal(they implement by distance maps), but our experiments show that such supervision is quite coarse and fragile, (do not recommend)
• Do not use the silhouette as the supervison, use differentiable rendering instead. You can refer to Softras and Neural Mesh Renderer for more details. The idea is to use mask(gt or by segmentation model) to supervise the shape of humans. But on datasets that only have a limited number of subjects(e.g. Human3.6M), such practice may not work.

Good luck!