search

hongsukchoi / Pose2Mesh_RELEASE

Official Pytorch implementation of "Pose2Mesh: Graph Convolutional Network for 3D Human Pose and Mesh Recovery from a 2D Human Pose", ECCV 2020
MIT License
677 stars 69 forks source link

Question about rescale_L #39

Closed SeanChenxy closed 2 years ago

SeanChenxy commented 2 years ago

https://github.com/hongsukchoi/Pose2Mesh_RELEASE/blob/e91bdd69077e964c5591f7c52b6f484d5180bd12/lib/coarsening.py#L32 Hi, thank you for sharing the code. I have a question about rescale_L. Why not use L /= lmax * 2 rather than L = 2.0*L / lmax? Looking forward your reply.

hongsukchoi commented 2 years ago

hi, i followed the code here: https://github.com/3d-hand-shape/hand-graph-cnn/blob/master/hand_shape_pose/util/coarsening.py

I think it doesn’t matter, since the purpose of rescaling is to prevent value explosion during chebyshev conv. With the current code, the eigenvalue of a laplacian matrix is still bounded in [-1,1]

SeanChenxy commented 2 years ago

Get it. Thanks a lot.