PyTorch-StudioGAN VS BigGAN-PyTorch

Given the first author I'd expect it to land in StudioGAN sometime in the future. Training it from scratch will definitely be costly though.

Github Link: https://github.com/POSTECH-CVLab/PyTorch-StudioGAN

Github Link: https://github.com/POSTECH-CVLab/PyTorch-StudioGAN Paper Link: https://arxiv.org/abs/2206.09479 I would like to introduce PyTorch-StudioGAN library, which I have been maintaining for the past two years. StudioGAN is a PyTorch library providing implementations of representative Generative Adversarial Networks (GANs) for conditional/unconditional image generation. StudioGAN aims to offer an identical playground for modern GANs so that machine learning researchers can readily compare and analyze a new idea. Moreover, StudioGAN provides an unprecedented-scale benchmark for generative models. The benchmark includes results from GANs (BigGAN-Deep, StyleGAN-XL), auto-regressive models (MaskGIT, RQ-Transformer), and Diffusion models (LSGM++, CLD-SGM, ADM-G-U). [Features] * Coverage: StudioGAN is a self-contained library that provides 7 GAN architectures, 9 conditioning methods, 4 adversarial losses, 13 regularization modules, 6 augmentation modules, 8 evaluation metrics, and 5 evaluation backbones. Among these configurations, we formulate 30 GANs as representatives. * Flexibility: Each modularized option is managed through a configuration system that works through a YAML file, so users can train a large combination of GANs by mix-matching distinct options. * Reproducibility: With StudioGAN, users can compare and debug various GANs with the unified computing environment without concerning about hidden details and tricks. * Plentifulness: StudioGAN provides a large collection of pre-trained GAN models, training logs, and evaluation results. * Versatility: StudioGAN supports 5 types of acceleration methods with synchronized batch normalization for training: a single GPU training, data-parallel training (DP), distributed data-parallel training (DDP), multi-node distributed data-parallel training (MDDP), and mixed-precision training.

Code for https://arxiv.org/abs/2111.01118 found: https://github.com/POSTECH-CVLab/PyTorch-StudioGAN

StudioGAN is introduced: A PyTorch library for SoTA GAN models

Github: https://github.com/POSTECH-CVLab/PyTorch-StudioGAN

The second link gives you the entire source code for training the model https://github.com/ajbrock/BigGAN-PyTorch/tree/master . Looks like BigGAN.py and BigGANdeep.py are the two files that define the architecture. Can you work with that?

BigGAN (https://github.com/ajbrock/BigGAN-PyTorch) - This is a PyTorch implementation of the BigGAN model for generating high-resolution images. It is trained on a large dataset and can generate a wide range of images, including photographs of animals, objects, and landscapes.

Tangentially, theoretically you can't use BN in the discriminator for WGAN-GP anyway (assuming that you're using the Gulrajani work) because it breaks the sample independence assumptions of the GP. If you have a relatively structured dataset (eg all faces, all cars, all giraffes, etc.) and no class conditioning, look into StyleGAN2-ADA for the best results. If you have a dataset with a lot of variation and a lot of classes try using the BigGAN repo.

BigGan και Deep Dream https://github.com/ajbrock/BigGAN-PyTorch https://github.com/google/deepdream