/ACGAN-PyTorch

Primary LanguagePythonMIT LicenseMIT

Conditional Image Synthesis With Auxiliary Classifier GANs

See README_og.md for complete details. Forked from https://github.com/gitlimlab/ACGAN-PyTorch

Example Usage:

For cpu only remove CUDA_VISIBLE_DEVICES=1 and --cuda.

For MNIST zeros and ones

Improved GAN

CUDA_VISIBLE_DEVICES=1 python main.py --outf=/scratch0/ilya/locDoc/ACGAN/experiments/julytest13 --train_batch_size=32 --cuda --dataset mnist_subset --num_classes 2 --imageSize=32 --data_root=/scratch0/ilya/locDoc/data/mnist --eval_period 50 --nc 1 --ndf 32 --ngf 32 --GAN_lrD 0.0001 --g_loss feature_matching

Mary GAN

CUDA_VISIBLE_DEVICES=1 python main.py --outf=/scratch0/ilya/locDoc/ACGAN/experiments/julytest13 --train_batch_size=32 --cuda --dataset mnist_subset --num_classes 2 --imageSize=32 --data_root=/scratch0/ilya/locDoc/data/mnist --eval_period 50 --nc 1 --ndf 32 --ngf 32 --GAN_lrD 0.0001

activation maximization gan

CUDA_VISIBLE_DEVICES=1 python main.py --outf=/scratch0/ilya/locDoc/ACGAN/experiments/julytest7 --train_batch_size=32 --cuda --dataset=mnist_subset --imageSize=32 --data_root=/scratch0/ilya/locDoc/data/mnist --eval_period 50 --nc 1 --num_classes 2 --ndf 32 --ngf 32 --GAN_lrD 0.0001 --g_loss activation_maximization

complement GAN

CUDA_VISIBLE_DEVICES=1 python main.py --outf=/scratch0/ilya/locDoc/ACGAN/experiments/julytest9 --train_batch_size=32 --cuda --dataset=mnist_subset --imageSize=32 --data_root=/scratch0/ilya/locDoc/data/mnist --eval_period 50 --nc 1 --num_classes 2 --ndf 32 --ngf 32 --GAN_lrD 0.0001 --g_loss crammer_singer_complement --g_loss_aux confuse --g_loss_aux_weight 0.33 --confuse_margin 1.0

Full MNIST

CUDA_VISIBLE_DEVICES=1 python main.py --outf=/scratch0/ilya/locDoc/ACGAN/experiments/julytest10 --train_batch_size=32 --cuda --dataset=mnist --imageSize=32 --data_root=/scratch0/ilya/locDoc/data/mnist --eval_period 50 --nc 1 --num_classes 10 --ndf 32 --ngf 32 --GAN_lrD 0.0001 --g_loss crammer_singer_complement --g_loss_aux confuse --g_loss_aux_weight 0.33 --confuse_margin 1.0