champ icon indicating copy to clipboard operation
champ copied to clipboard
verified publisher icon fudan-generative-vision

Reproduced results of TikTok data set

Open skhu101 opened this issue 1 year ago • 2 comments

Thanks for your wonderful work. I follow your data processing script and training scripts to re-train a model on the TikTok data set. I have several questions. 1). The fitted SMPL result is somehow not very accurate using your data processing scripts. 2). The quality of visualization results of stages 1 and 2 are far from the results shown in your paper. I attached the SMPL results and visualization results from stages 1 and 2. Could you give me some suggestions? Looking forward to your reply. 0176 100000-val_0

https://github.com/user-attachments/assets/d9813718-59eb-4d31-a5c2-7f0e7deba601

skhu101 avatar Aug 07 '24 08:08 skhu101

I meet same issue in stage 1. The same issue in stage2 is solved by training enough steps for stage1 and stage2, however the issue in stage 1 is still remained.

SzhangS avatar Sep 12 '24 07:09 SzhangS

Could you please share the YAML file used for training? I am wondering if the issue might be caused by a misconfiguration of the aug_type parameter in the YAML file.

NinoNeumann avatar Oct 11 '24 08:10 NinoNeumann