search

Vegetebird / MHFormer

[CVPR 2022] MHFormer: Multi-Hypothesis Transformer for 3D Human Pose Estimation
MIT License
538 stars 85 forks source link

How to understand hypotheses? #5

Closed hellojialee closed 2 years ago

hellojialee commented 2 years ago

Hi, Thank you for your great work! How to understand that multiple hypotheses are truly reasonably predicted within the end-to-end model?

Vegetebird commented 2 years ago

Hi~Sorry for the late reply.

Different from existing methods that generate multiple hypotheses (one-to-many mapping), we argue that it is more reasonable to provide insight into how to improve the entire network architecture to be inherently more strong for the ambiguous inverse problem of 3D human pose estimation. We use multi-hypothesis representations only as intermediate variants to increase the diversity of the network (one-to-one mapping).