Request for MPI-INF-3DHP data preprocessing and evaluation code – unable to reproduce reported results

[llj110]

Thank you for the excellent work and for open-sourcing the code!
I am trying to reproduce the results reported in the paper on the MPI-INF-3DHP dataset, but I have encountered significant difficulties. Despite following the instructions carefully and trying multiple configurations, I am consistently unable to match the performance numbers reported in the paper (e.g., PCK and AUC metrics).

Could you please provide:
The exact data preprocessing code for MPI-INF-3DHP.
The official evaluation / testing script for MPI-INF-3DHP.

[xiu-cs]

Hi @llj110, thank you for your message and for your interest in our work.

For evaluation on MPI-INF-3DHP, I haven’t had a chance to organize the corresponding code yet. However, please note that we should use the normalized 3D pose annotations (universal), which are scaled according to the height of the universal skeleton used by Human3.6M. You can find the corresponding data processing in P-STMO, and 3DHP also explains this difference.

May I ask if you could send me your current processing code by email (tiwang.cs@gmail.com or ti.wang@epfl.ch)? Then we can look into the problem together and try to identify and solve the issue. Thanks!