Worse results on custom datasets THuman2.0

[JanaldoChen]

Thanks for your great work! When I run your code on a custom dataset THuman2.0(https://github.com/ytrock/THuman2.0-Dataset), I get very bad results. The THuman2.0 dataset contains 525 3d human scans, I use these 3d meshes to render multi-view images (from 65 virtual cameras in the hemisphere) and obtain smpl parameters using Easymocap(https://github.com/zju3dv/EasyMocap), and I process the new dataset following zju_mocap's format. During the experiment, I set time_steps=1 and the number of training_view is 4, but I got very worse results. Can you give me some suggestions? Or provide the script code to process custom data?