leoxiaobin / deep-high-resolution-net.pytorch

The project is an official implementation of our CVPR2019 paper "Deep High-Resolution Representation Learning for Human Pose Estimation"
https://jingdongwang2017.github.io/Projects/HRNet/PoseEstimation.html
MIT License
4.32k stars 910 forks source link

Failed to reproduce one result. #218

Open ShuangjunLiu opened 4 years ago

ShuangjunLiu commented 4 years ago

Hello, I can't reproduce result for setting "res101_256x256_d256x3_adam_lr1e-3"

mpii pckh0.5: Paper report : 89.6 %
From pretrained model (provided in this repo): 89.2%
Retrain it: 88.87%

Is there anything I missed I should pay attention to to retrain a model with reported performance?