Using progressive gorwing of GANs to augmentation skin lesion dataset.
This project was originated as SFU CMPT-726 (18 Fall) course project. Collaborator: Yongjun He, Ruochen Jiang.
I wrote an article (in Chinese) to discuss some tricks and implementation details. Zhihu article
Be careful to use the code!
- This may not be full implementation of progressive GANs, and we haven't run the code on that same dataset as the original paper.
- Unsolved issue: the training fails when using PyTorch 1.0. I'm still not sure what difference between 1.0 and 0.4 causes this issue...
- PyTorch 0.4 with torchvision
- tensorboardX
- Pillow
ISIC 2018 classification task.
python main.py --mode train
The training taks about 4 days on 4 NVIDIA P100 Pascal GPUs.
python main.py --mode test --num 1000
- tkarras/progressive_growing_of_gans (official Tensorflow implementation)
- akanimax/pro_gan_pytorch
- nashory/pggan-pytorch