Realtime_Multi-Person_Pose_Estimation
Realtime_Multi-Person_Pose_Estimation copied to clipboard
Limb width is hardcoded?
The images in the paper seem to indicate that the limb width is precisely labeled?
However, looking at the code (function putVecMaps):
https://github.com/CMU-Perceptual-Computing-Lab/caffe_train/blob/master/src/caffe/cpm_data_transformer.cpp
thre seems to be the limb width, and it's hardcoded to 1. isn't 1 too small for limb width? why was one chosen as the width?