search

HRNet / HigherHRNet-Human-Pose-Estimation

This is an official implementation of our CVPR 2020 paper "HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation" (https://arxiv.org/abs/1908.10357)
MIT License
1.35k stars 272 forks source link

The adjustment of keypoint position #67

Open pdoublerainbow opened 4 years ago

pdoublerainbow commented 4 years ago

Hi, author. I have a question about the code in keypoint position adjustment part. The code is here https://github.com/HRNet/HigherHRNet-Human-Pose-Estimation/blob/master/lib/core/group.py#L195, the question is that why the position value need to add 0.5 .

bowenc0221 commented 4 years ago

This is because during training keypoints are rounded to integer coordinate (but their location could be continuous value). A 0.5 is added here to compensate the quantization error.

seekFire commented 3 years ago

@bowenc0221 How is the value(0.5) determined? Is it because the size of final_heatmaps is 0.5X that of original image ?