wangziren1

您好，我看了你的项目，做的很棒，我也想要你的数据集，可以发给我一份吗？[email protected]

paper name:Pose Proposal Networks

mpi_annotation.json is generated by [openpose repo](https://github.com/ZheC/Realtime_Multi-Person_Pose_Estimation). The code should be in a certain issue. Because of long time, I don't know which issue exactly. You can try to search some...

> ？ Sorry, that is misoperation. Next, I will answer your question. 1.Input videos has different resolution, so they need to be processed to have same size as network input....

Maybe, dropping learning rate by ten times or five times helps.