Image Reconstruction on COCO 2017 (val)
33.6PSNRCVVAE-SD3
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CVVAE-SD3Params=68M + 114M, FCR=4x, Comp.=true2024.05 | 33.6 | 0.916 | 0.072 | — | — | |
| VAE-SD3Params=34M + 49M, FCR=1x2024.05 | 32.6 | 0.899 | 0.064 | — | — | |
| LBI-GDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 32.52 | 0.9211 | 0.0369 | — | — | |
| LBI-HDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 32.48 | 0.9207 | 0.0377 | — | — | |
| LBI-NDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 32.41 | 0.9203 | 0.0373 | — | — | |
| GQbpp (K x N)=1.00 (2^16 x 4096), Backbone=UNet based2025.12 | 32.36 | 0.915 | 0.02 | — | 1.875 | |
| LV-RAETokenizer Config=f16d7682026.02 | 32.32 | 0.941 | 0.015 | 13.2 | — | |
| FSQbpp (K x N)=1.00 (2^16 x 4096), Backbone=UNet based2025.12 | 32.3 | 0.917 | 0.022 | — | 2.797 | |
| VQAE (upper bound)Diffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 32.28 | 0.8834 | 0.0407 | — | — | |
| VQGANbpp (K x N)=1.00 (2^16 x 4096), Backbone=UNet based2025.12 | 31.97 | 0.901 | 0.024 | — | 3.455 | |
| BELMDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=64 K2026.03 | 31.9 | 0.8773 | 0.0437 | — | — | |
| EDICTDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=64 K2026.03 | 31.84 | 0.8757 | 0.0482 | — | — | |
| AIDI-ADiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 31.71 | 0.8784 | 0.0475 | — | — | |
| AIDI-EDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 31.67 | 0.8785 | 0.0477 | — | — | |
| IFENFE=50, Backbone=Stable Diffusion v1.42025.12 | 31.658 | 0.733 | 0.216 | — | — | |
| LBI-H (w/o ILB)Diffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 31.61 | 0.8777 | 0.0482 | — | — | |
| FPIDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 31.6 | 0.8777 | 0.048 | — | — | |
| LBI-G (w/o ILB)Diffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 31.58 | 0.8782 | 0.0485 | — | — | |
| GQbpp (K x N)=1.00 (2^16 x 4096), Backbone=ViT based2025.12 | 31.5 | 0.908 | 0.022 | — | 2.703 | |
| FSQbpp (K x N)=1.00 (2^16 x 4096), Backbone=ViT based2025.12 | 31.48 | 0.911 | 0.023 | — | 3.045 | |
| LBI-N (w/o ILB)Diffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 31.48 | 0.8762 | 0.0498 | — | — | |
| TICDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=10.88 G2026.03 | 31.47 | 0.8678 | 0.0622 | — | — | |
| BSQbpp (K x N)=1.00 (2^16 x 4096), Backbone=ViT based2025.12 | 31.38 | 0.918 | 0.026 | — | 2.835 | |
| VQGANbpp (K x N)=1.00 (2^16 x 4096), Backbone=ViT based2025.12 | 31.07 | 0.904 | 0.029 | — | 3.494 | |
| ReNoiseNFE=150, Backbone=Stable Diffusion v1.42025.12 | 31.025 | 0.641 | 0.316 | — | — | |
| LV-RAE + Noise AugmentationTokenizer Config=f16d768, Noise Augmentation=true2026.02 | 30.91 | 0.93 | 0.018 | 8.74 | — | |
| FLUX-VAETokenizer Config=f8d162026.02 | 30.89 | 0.921 | 0.015 | 8.86 | — | |
| LBI-HDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 30.72 | 0.8872 | 0.0327 | — | — | |
| LBI-GDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 30.65 | 0.8869 | 0.0336 | — | — | |
| LBI-NDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 30.57 | 0.885 | 0.0353 | — | — | |
| BSQbpp (K x N)=1.00 (2^16 x 4096), Backbone=UNet based2025.12 | 30.33 | 0.906 | 0.031 | — | 2.638 | |
| ExactDPMDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 30.33 | 0.8563 | 0.1281 | — | — | |
| EasyInvNFE=50, Backbone=Stable Diffusion v1.42025.12 | 30.189 | 0.646 | 0.321 | — | — | |
| GQbpp (K x N)=0.50 (2^16 x 2048), Backbone=ViT based2025.12 | 30.18 | 0.887 | 0.034 | — | 3.616 | |
| GQbpp (K x N)=0.50 (2^16 x 2048), Backbone=UNet based2025.12 | 30.14 | 0.877 | 0.037 | — | 3.116 | |
| PTIDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=7.58 M2026.03 | 30.11 | 0.8629 | 0.0695 | — | — | |
| VQAE (upper bound)Diffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 30.09 | 0.8368 | 0.0423 | — | — | |
| ExactDPMDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 30.08 | 0.8504 | 0.0942 | — | — | |
| ReNoiseDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=64 K2026.03 | 29.76 | 0.8561 | 0.0805 | — | — | |
| DDIM InversionNFE=50, Backbone=Stable Diffusion v1.42025.12 | 29.717 | 0.621 | 0.328 | — | — | |
| LBI-H (w/o ILB)Diffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 29.65 | 0.8318 | 0.0493 | — | — | |
| LBI-G (w/o ILB)Diffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 29.62 | 0.832 | 0.0497 | — | — | |
| LBI-N (w/o ILB)Diffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 29.57 | 0.8307 | 0.0508 | — | — | |
| AIDINFE=140, Backbone=Stable Diffusion v1.42025.12 | 29.107 | 0.563 | 0.373 | — | — | |
| FSQbpp (K x N)=0.50 (2^16 x 2048), Backbone=UNet based2025.12 | 29.08 | 0.855 | 0.043 | — | 4.008 | |
| VQGANbpp (K x N)=0.50 (2^16 x 2048), Backbone=UNet based2025.12 | 29.06 | 0.839 | 0.049 | — | 6.616 | |
| ExactDPM (w/o DI)Diffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 28.56 | 0.8242 | 0.0551 | — | — | |
| FSQbpp (K x N)=0.50 (2^16 x 2048), Backbone=ViT based2025.12 | 28.51 | 0.851 | 0.053 | — | 5.39 | |
| ExactDPM (w/o DI)Diffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 28.28 | 0.8187 | 0.1368 | — | — | |
| BSQbpp (K x N)=0.50 (2^16 x 2048), Backbone=ViT based2025.12 | 28.19 | 0.858 | 0.049 | — | 4.587 | |
| LFQbpp (K x N)=1.00 (2^16 x 4096), Backbone=UNet based2025.12 | 28.16 | 0.845 | 0.072 | — | 11.121 | |
| LBI-GDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 28.14 | 0.8338 | 0.0332 | — | — | |
| LBI-NDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 28.12 | 0.8328 | 0.0344 | — | — | |
| LBI-HDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 28.04 | 0.831 | 0.0372 | — | — | |
| VQGANbpp (K x N)=0.50 (2^16 x 2048), Backbone=ViT based2025.12 | 27.83 | 0.832 | 0.058 | — | 5.461 | |
| CV-VAE (3D)Params=100M + 156M, Frame Compression Rate (FCR)=4x, Compatibility with generative models=true2024.05 | 27.7 | 0.805 | 0.135 | — | — | |
| EDICTDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=36 K2026.03 | 27.68 | 0.8053 | 0.0927 | — | — | |
| CV-VAE (2D+3D)Params=68M + 114M, Frame Compression Rate (FCR)=4x, Compatibility with generative models=true2024.05 | 27.6 | 0.805 | 0.136 | — | — | |
| BSQbpp (K x N)=0.50 (2^16 x 2048), Backbone=UNet based2025.12 | 27.58 | 0.844 | 0.057 | — | 4.465 | |
| GQbpp (K x N)=0.25 (2^16 x 1024), Backbone=ViT based2025.12 | 27.55 | 0.83 | 0.06 | — | 5.305 | |
| LFQbpp (K x N)=0.50 (2^16 x 2048), Backbone=ViT based2025.12 | 27.54 | 0.833 | 0.067 | — | 8.7 | |
| TICDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=1.67 G2026.03 | 27.51 | 0.8112 | 0.0591 | — | — | |
| PTIDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=3.76 M2026.03 | 27.5 | 0.8059 | 0.0768 | — | — | |
| ReNoiseDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=36 K2026.03 | 27.48 | 0.8094 | 0.0821 | — | — | |
| RIVALDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=2.42 G2026.03 | 27.39 | 0.8131 | 0.0612 | — | — | |
| GQbpp (K x N)=0.25 (2^16 x 1024), Backbone=UNet based2025.12 | 27.29 | 0.816 | 0.057 | — | 3.797 | |
| ExactDPMDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 27.29 | 0.7791 | 0.1989 | — | — | |
| VAE-OSPParams=94M + 135M, Frame Compression Rate (FCR)=4x, Compatibility with generative models=false2024.05 | 27 | 0.791 | 0.142 | — | — | |
| BDIADiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=64 K2026.03 | 26.96 | 0.832 | 0.0936 | — | — | |
| AIDI-EDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 26.87 | 0.777 | 0.1441 | — | — | |
| AIDI-ADiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 26.85 | 0.7765 | 0.1446 | — | — | |
| FPIDiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=02026.03 | 26.83 | 0.7765 | 0.1447 | — | — | |
| VQAE (upper bound)Diffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 26.75 | 0.7699 | 0.0512 | — | — | |
| BELMDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=16 K2026.03 | 26.71 | 0.7688 | 0.0518 | — | — | |
| TICDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=759.38 M2026.03 | 26.69 | 0.7684 | 0.0546 | — | — | |
| LBI-N (w/o ILB)Diffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 26.65 | 0.768 | 0.0547 | — | — | |
| BDIADiffusion Model Variant=Stable Diffusion v2.1 (SD2), Additional Parameters=36 K2026.03 | 26.65 | 0.8035 | 0.0898 | — | — | |
| LBI-G (w/o ILB)Diffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 26.63 | 0.7691 | 0.0557 | — | — | |
| NTIDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=7.58 M2026.03 | 26.61 | 0.834 | 0.1077 | — | — | |
| VAE-SD2.1Params=34M + 49M, Frame Compression Rate (FCR)=1x, Compatibility with generative models=N/A2024.05 | 26.6 | 0.773 | 0.127 | — | — | |
| FPIDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 26.59 | 0.7671 | 0.055 | — | — | |
| LBI-H (w/o ILB)Diffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 26.54 | 0.7663 | 0.056 | — | — | |
| LFQbpp (K x N)=0.50 (2^16 x 2048), Backbone=UNet based2025.12 | 26.47 | 0.805 | 0.103 | — | 17.508 | |
| AIDI-ADiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 26.38 | 0.7595 | 0.0624 | — | — | |
| LFQbpp (K x N)=1.00 (2^16 x 4096), Backbone=ViT based2025.12 | 26.36 | 0.794 | 0.103 | — | 20.381 | |
| ExactDPM (w/o DI)Diffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 26.27 | 0.7603 | 0.0667 | — | — | |
| VQGANbpp (K x N)=0.25 (2^16 x 1024), Backbone=UNet based2025.12 | 26.25 | 0.756 | 0.099 | — | 14.11 | |
| EDICTDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=16 K2026.03 | 26.18 | 0.7581 | 0.0723 | — | — | |
| BSQbpp (K x N)=0.25 (2^16 x 1024), Backbone=ViT based2025.12 | 26.15 | 0.798 | 0.082 | — | 7.034 | |
| ReNoiseDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=16 K2026.03 | 26.12 | 0.7593 | 0.066 | — | — | |
| PTIDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=2.81 M2026.03 | 26.11 | 0.7581 | 0.0747 | — | — | |
| VA-VAETokenizer Config=f16d322026.02 | 26.07 | 0.799 | 0.044 | 18.76 | — | |
| GNRIDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 26.05 | 0.8129 | 0.1501 | — | — | |
| FSQbpp (K x N)=0.25 (2^16 x 1024), Backbone=UNet based2025.12 | 26.01 | 0.767 | 0.072 | — | 5.451 | |
| Λ24-SQ-ViTArch.=T, Quant.=Λ24-SQ, Param.=174M, #bits=18, TP=45.12025.12 | 26 | 0.8008 | 0.0632 | 5.15 | — | |
| AIDI-EDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=02026.03 | 25.98 | 0.7565 | 0.0803 | — | — | |
| NPIDiffusion Model Variant=Stable Diffusion XL v1.0 (SDXL), Additional Parameters=02026.03 | 25.93 | 0.8065 | 0.1542 | — | — | |
| FSQbpp (K x N)=0.25 (2^16 x 1024), Backbone=ViT based2025.12 | 25.85 | 0.765 | 0.112 | — | 11.213 | |
| NTIDiffusion Model Variant=Stable Diffusion v1.5 (SD1), Additional Parameters=2.81 M2026.03 | 25.81 | 0.7566 | 0.0765 | — | — | |
| SD-VAETokenizer Config=f8d42026.02 | 25.38 | 0.76 | 0.065 | 43.72 | — |