Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models
Open paper
Benchmark
Model
Rank
Results
image-compression-on-imagenet-256x256-vae-reconstruction-f16
VA-VAE-f16d32 (DINOv2)
#3
PSNR: 27.96
SSIM: 0.79
image-generation-on-imagenet-256x256
LightningDiT + VA-VAE (with the guidance interval)
#21
FID: 1.35
Rank counts only results with a code link.