Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Does use_gt_pelvis correspond to MPJPE relative to pelvis in the paper?

See original GitHub issue

Does use_gt_pelvis correspond to MPJPE relative to pelvis in the paper? If so, why are the results worse with GT pelvis? Thanks!

Issue Analytics

State:
Created 4 years ago
Comments:6 (1 by maintainers)

Top GitHub Comments

1reaction

yurymalkovcommented, Nov 13, 2019

@yihui-he The multi-view method always estimates the positions in world coordinates for all of the joints, including the pelvis. MPJPE relative to pelvis is the most common metric, so for a fair comparison in Table 1 we subtract the position of the pelvis from all joints. We argue that for multi-view methods it makes more sense not to subtract the pelvis (as we did in Table 2).

0reactions

yihui-hecommented, Nov 13, 2019

Thanks for the explanation

Top Results From Across the Web

learnable-triangulation-pytorch/README.md at master - GitHub

This repository is an official PyTorch implementation of the paper "Learnable Triangulation of ... Precalculated results, MPJPE (relative to pelvis), mm ...