Clone the package probvis, which is needed to visualize the results. Then create a conda environment as follows:
conda env create -f ./environments/env_macos.yml
and you should be ready to to go.
-
ld
: Parameter that controls the shape of the non-uniform distribution, namely$\lambda_{dist}$ -
lp
: Parameter that controls the percentage of the non-uniform samples per mini-bartch, namely$\lambda_{perc}$ -
lr
: Parameter that controls reconstruction regularization, namely$\lambda_{cyc}$ -
gan_type
: Select the model to train: mdgan, pmdgan, epmdgan.
python3 main.py --n_epochs 800 --z_dim 256 --dataset c10 --data_path ../Data --gan_type epmdgan --gpu 0 --ckpt_dir experiments/c10/3 --lr 3 --lp 0.8 --ld 4
This code uses the folowing repositories:
-
Computing the FID score: code
Pablo Sanchez - For any questions, comments or help to get it to run, please don't hesitate to mail me: pablo.sanchez-martin@tuebingen.mpg.de