search

ilovepose / DarkPose

Distribution-Aware Coordinate Representation for Human Pose Estimation
https://ilovepose.github.io/coco
Apache License 2.0
550 stars 80 forks source link

Question about a small detail #26

Closed seekFire closed 3 years ago

seekFire commented 3 years ago

In evaluate.py, function calc_dists, the Euclidean distance will be calculated under the condition:

if target[n, c, 0] > 1 and target[n, c, 1] > 1:

It seems that you exclude the case where the target coordinates are in [0, 1], so why do this?

seekFire commented 3 years ago

I think you may aim to preserve the groundtruth coordinates available after carrying on one pixel shifting of flip testing. Is it right?

xizero00 commented 3 years ago

@seekFire You may need to ask the author of HRNet https://github.com/leoxiaobin/deep-high-resolution-net.pytorch/