Experimental setup of the outfit-specific version
tpzou opened this issue · comments
Hi @jsnln , congrats on this great work, and thanks a lot for open-sourcing this project!
I note that the experimental setup of the outfit-specific version was not detailed in the article. Could you describe it specifically? Such as training split and testing split of dataset?
Thanks for your interest in our work. To run an outfit specific version, simply put a single outfit in the config file xxx_subject_list.yaml
where xxx
is the name of the experiment. Note that you should also specify the outfit in step1.yaml
and step2.yaml
.
The training and testing split are the official splits of ReSynth (https://pop.is.tue.mpg.de/index.html)
Thanks for your quick reply! The per-outfit experiment in POP looks very similar to the outfit-specific version. But I still can‘t find the detailed information in the code or paper.
As there are 37 motion sequences of 03375_blazerlong in CAPE, which sequences do you use for the test set and which ones for the training set? Or is there another split way?
In the download page of the CAPE (https://cape.is.tue.mpg.de/download.php) you can find a version of CAPE packed for POP, which defines a train/test split. We used this official split.