junhocho / SRGAN

Implementation of [Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network arXiv:1609.04802v2]
MIT License
82 stars 19 forks source link

How to train using other datasets instand of imagenet? #4

Open zhangqianhui opened 7 years ago

zhangqianhui commented 7 years ago

Imagenet is very big and How to use celebA or other dataset to train SRGAN? Can you help me ?

junhocho commented 7 years ago

If you see ./src/util.lua, prepImageNet function simply goes to ImageNet directory and count and read all file names. Just change line 70~80 in ./train-SRResNet.lua into CelebA or something. Fix prepImageNet as you wish because the dataset directory structure is likely to be different.

zhangqianhui commented 7 years ago

thanks