lucidrains/lightweight-gan

Health v2 · maintenance only

Adoption not part of the score

1682 stars · 219 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A PyTorch implementation of the 'lightweight' GAN from ICLR 2021, featuring skip-layer excitation in the generator and autoencoding self-supervised learning in the discriminator. It enables high-resolution image generation (up to 1024px) trainable on a single GPU within hours to a couple of days.

Use cases

• train a GAN to generate high-resolution images from a small dataset
• generate 512x512 images on a single GPU in hours
• generate images from a few hundred training photos
• experiment with data augmentation for low-data GAN training
• preview augmentation effects on images before training

When to choose

• you need to train an image generator quickly on limited hardware
• you have a small image dataset (sub-hundred images) and want high-resolution output
• you want a simple one-command GAN training pipeline in PyTorch

When to avoid

• you need state-of-the-art diffusion-model quality rather than GANs
• you need text-to-image or conditional generation
• you require large-scale distributed training infrastructure

Facets

library · maturity maintenance

machine-learning deep-learning image-processing cli artificial-intelligence deep-learning image-processing python cli gan generative-adversarial-network pytorch image-generation iclr-2021 gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="lucidrains/lightweight-gan")