ConceptVQ-GAN
VQGAN
VQGAN is an image encoding model that combines vector quantization with adversarial training to compress images efficiently into discrete token sequences. Its reconstruction quality surpasses that of standard VQ-VAE, and it is commonly used in image generation pipelines and visual autoregressive models.
Timeline (last 90 days)
Oct 9
The lightweight tokenizer direction is inspired by visual autoregressive models (such as VQVAE and VQGAN), encoding images into discrete token sequences through vector quantization
Unverified50%