Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models

Benchmark Model Rank Results
image-compression-on-imagenet-256x256-vae-reconstruction-f16VA-VAE-f16d32 (DINOv2)#3PSNR: 27.96SSIM: 0.79
image-generation-on-imagenet-256x256LightningDiT + VA-VAE (with the guidance interval)#21FID: 1.35