GAN and VAE comparisons

Generating samples from model

Example command for gan:

python gan.py sample --model_path runs/saved_models/best_gan.pt --n_samples 1000 --samples_dir results/samples/gan/images

Example command for vae:

python vae.py sample --model_path runs/saved_models/best_vae.pt --n_samples 1000 --samples_dir results/samples/vae/images

(Above commands used for generating samples used in FID calculation)

Quantitative evaluations

Use score_fid.py script for calculating FID between samples from svhn testset and model-generated samples saved in folder.

Thousand samples generated from the DCGAN and VAE model implemented here (trained on svhn trainset) are saved in results/samples/gan/images/ and results/samples/vae/images/ respectively.

Qualitative evaluations

main.py is written for running qualitative evaluations of the trained gan and vae models.

Reference

DCGAN Paper