silleellie / painter_gan Goto Github PK

In case something has been implemented wrongly

The dataset that now all GANs architecture must use is the following: Paintings dataset

Perform data augmentation by giving more importance to impressionism - post impressionism paintings, but use all paintings style

The __main__ method of each GAN implementation contains the whole experiment to be executed, but there are some parts in common: design a better abstraction by considering the implementation of the GAN abstract class

The metrics to consider are MFID, FID, Inception Score, etc.

To perform a fairer comparison, n° epoch and batch size should be equal accross each GAN implementation, the hyperparameter to tune are surely the dimension of the latent noise vector, and n° of filters.
After that other hyperparameters, typical of each GAN implementation, can be considered