This is an official implementation of our CVPR 2022 ORAL paper "Temporal Feature Alignment and Mutual Information Maximization for Video-Based Human Pose Estimation" .
With the source code here, I can only reach the level of mAP 82.2% in the PoseTrack2017 validation set,but it is 84.8% in the author's paper (three rounds of experiments have been done at present. Only PoseTrack2017 training set was used during training). Can anybody reach mAP 84.8% in PoseTrack2017 validation set with this repository?
With the source code here, I can only reach the level of mAP 82.2% in the PoseTrack2017 validation set,but it is 84.8% in the author's paper (three rounds of experiments have been done at present. Only PoseTrack2017 training set was used during training). Can anybody reach mAP 84.8% in PoseTrack2017 validation set with this repository?