search

XingangPan / GAN2Shape

Code for GAN2Shape (ICLR2021 oral)
https://arxiv.org/abs/2011.00844
MIT License
571 stars 104 forks source link

About pre-training #2

Closed 07hyx06 closed 3 years ago

07hyx06 commented 3 years ago

Thanks for your great work!

What about the results of pre-training on the whole dataset (e.g. CelebA) and test on one unseen image without any optimization? Just like the setting in Unsup3D.

XingangPan commented 3 years ago

@07hyx06 I haven't tried that. But if we pre-train on only 200 samples and test on an unseen image without optimization, the result would be coarse without details. I guess pre-training on the whole dataset would get better results, but not as precise as the result with instance-specific optimization.