search

cuiaiyu / dressing-in-order

(ICCV'21) Official code of "Dressing in Order: Recurrent Person Image Generation for Pose Transfer, Virtual Try-on and Outfit Editing" by Aiyu Cui, Daniel McKee and Svetlana Lazebnik
https://cuiaiyu.github.io/dressing-in-order
Other
507 stars 126 forks source link

How to Train on High Resolution #53

Closed NorthanX closed 2 years ago

NorthanX commented 2 years ago

Hi, thanks for the great work! I am curious about the results with high resolution, like 512*358, so I want to train by my own, but I am not so sure about how to change the training process. Could you please give me some advice?

cuiaiyu commented 2 years ago

Hi!

You can retrain the model with input in 512 by specifying --load_size 512 or --crop_size 512 (I forget which one). If you get any error saying dimension doesn't match, please fix the hard code (which gives the error) from 256 to the input argument opt.load_size, otherwise the code should be runnable.

GFLA flownet is likely requiring some retraining to 512 as well, because GFLA is also trained at 256x256. I am not sure how robust GFLA is with higher input resolution.

I'm not sure if the performance would maintain the same, as this work is only tested at 256. Some related work in higher resolution that may be interesting are:

https://pose-with-style.github.io/ https://tryongan.github.io/tryongan/

NorthanX commented 2 years ago

Thank you for your advice! I will try with that.