ConceptVQ-VAE / Vector Quantised-Variational AutoEncoder
VQVAE
VQ-VAE (Vector Quantized Variational Autoencoder) is an autoencoder model that discretizes continuous latent representations through vector quantization, encoding images into sequences of discrete tokens. It is widely applied in image generation and multimodal unified architectures that require discrete latent spaces.
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%