<ul><li>A novel continuous-space latent diffusion framework called VQ-LCMD is introduced, allowing generative modeling of discrete data.</li><li>VQ-LCMD combines joint embedding-diffusion training with a consistency-matching (CM) loss to stabilize training and enhance performance.</li><li>Experiments demonstrate that VQ-LCMD outperforms discrete-state latent diffusion models on FFHQ, LSUN Churches, and LSUN Bedrooms benchmarks.</li><li>VQ-LCMD achieves a FID of 6.81 for class-conditional image generation on ImageNet with 50 steps.</li></ul>

Improving Vector-Quantized Image Modeling with Latent Consistency-Matching Diffusion

Discover more