search

microsoft / human-pose-estimation.pytorch

The project is an official implement of our ECCV2018 paper "Simple Baselines for Human Pose Estimation and Tracking(https://arxiv.org/abs/1804.06208)"
MIT License
2.93k stars 601 forks source link

MPII Test Set Size Mismatch #106

Open NiteshBharadwaj opened 5 years ago

NiteshBharadwaj commented 5 years ago

Your test.json contains 7247 test samples while the original .mat file contains 11731 test samples. Can you explain the discrepancy? In effect, can you tell me how to start from this codebase and make a submission to leader board for single person pose?

HuangChenyang123 commented 5 years ago

Have you solve this problem? I meet the same problem.

xtyDoge commented 5 years ago

@HuangChenyang123 Maybe you can see https://github.com/anibali/eval-mpii-pose, it provides a test.h5 file that contains 11731 individuals. You can convert it to json file and use it to replace test.json to get the pred.mat in dimension of [11731,16,2]. And eval-mpii-pose also provides tools for convert this pred.mat to the format that MPII website required.

NiteshBharadwaj commented 5 years ago

I used the same method as described by @xtyDoge

HuangChenyang123 commented 5 years ago

@xtyDoge I don't know how to convert the test.h5 file to json file. I would appreciate it if you could send me the json file. My email address is 464383828@qq.com ,Thank you!

HuangChenyang123 commented 5 years ago

@NiteshBharadwaj I don't know how to convert the test.h5 file to json file. I would appreciate it if you could send me the json file. My email address is 464383828@qq.com ,Thank you!!!