aandyw/AnimeGen

An attempt to generate industry-standard facial images for anime characters using Generative Adversarial Networks

AnimeGen V1

An attempt to generate industry-standard facial images for anime characters using Generative Adversarial Networks

Features

• Self Attention Module + hinge loss
• Conditional Batch Normalization
• Big GAN + residual blocks
• Instance Noise
• Tensorboard visualization

Results

SAGAN V2 Epoch 217

SAGAN with INSTANCE NOISE

Losses

SAGAN V1	SAGAN V2

SAGAN with INSTANCE NOISE

Prerequisites

• Python 3.7.3
• Pytorch 1.2.0
• Numpy 1.17.2

Todo

• BCE Loss
• Tagging Data
• Get Big GAN to work?