Image Compression on CelebA (test)
33.77PSNROAMP+BM3D
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OAMP+BM3DCompression ratio (beta)=0.7, AWGN noise level (sigma)=0.052026.04 | 33.77 | — | 91.6 | |
| Diffusion-OAMP (DDIM)Compression ratio (beta)=0.7, AWGN noise level (sigma)=0.052026.04 | 33.34 | — | 89.3 | |
| Diffusion-OAMP (FM)Compression ratio (beta)=0.7, AWGN noise level (sigma)=0.052026.04 | 33.18 | — | 93.26 | |
| DDRMCompression ratio (beta)=0.7, AWGN noise level (sigma)=0.05, NFEs=502026.04 | 32.88 | — | 91.07 | |
| Diffusion-OAMP (DDIM)Compression ratio (beta)=0.4, AWGN noise level (sigma)=0.052026.04 | 32.62 | — | 89.52 | |
| OAMP+BM3DCompression ratio (beta)=0.4, AWGN noise level (sigma)=0.052026.04 | 32.05 | — | 88.54 | |
| DDRMCompression ratio (beta)=0.7, AWGN noise level (sigma)=0.05, NFEs=202026.04 | 31.97 | — | 89.78 | |
| Diffusion-OAMP (FM)Compression ratio (beta)=0.4, AWGN noise level (sigma)=0.052026.04 | 31.69 | — | 91.25 | |
| Diffusion-OAMP (DDIM)Compression ratio (beta)=0.1, AWGN noise level (sigma)=0.052026.04 | 30.3 | — | 86.4 | |
| DDRMCompression ratio (beta)=0.4, AWGN noise level (sigma)=0.05, NFEs=502026.04 | 30.17 | — | 86.29 | |
| Diffusion-OAMP (FM)Compression ratio (beta)=0.1, AWGN noise level (sigma)=0.052026.04 | 28.38 | — | 84.45 | |
| DDRMCompression ratio (beta)=0.4, AWGN noise level (sigma)=0.05, NFEs=202026.04 | 28.16 | — | 82.59 | |
| OAMP+BM3DCompression ratio (beta)=0.1, AWGN noise level (sigma)=0.052026.04 | 28.08 | — | 78.56 | |
| Diffusion-OAMP (DDIM)Compression ratio (beta)=0.7, AWGN noise level (sigma)=0.52026.04 | 25.99 | — | 76.45 | |
| DDRMCompression ratio (beta)=0.7, AWGN noise level (sigma)=0.5, NFEs=502026.04 | 24.85 | — | 71.31 | |
| Diffusion-OAMP (DDIM)Compression ratio (beta)=0.4, AWGN noise level (sigma)=0.52026.04 | 24.66 | — | 72.35 | |
| DDRMCompression ratio (beta)=0.7, AWGN noise level (sigma)=0.5, NFEs=202026.04 | 24.5 | — | 70.25 | |
| Diffusion-OAMP (FM)Compression ratio (beta)=0.7, AWGN noise level (sigma)=0.52026.04 | 23.92 | — | 70.66 | |
| DDRMCompression ratio (beta)=0.4, AWGN noise level (sigma)=0.5, NFEs=502026.04 | 23.24 | — | 65.75 | |
| OAMP+BM3DCompression ratio (beta)=0.7, AWGN noise level (sigma)=0.52026.04 | 23.05 | — | 59.96 | |
| Diffusion-OAMP (FM)Compression ratio (beta)=0.4, AWGN noise level (sigma)=0.52026.04 | 22.83 | — | 65.01 | |
| DDRMCompression ratio (beta)=0.4, AWGN noise level (sigma)=0.5, NFEs=202026.04 | 22.32 | — | 63.21 | |
| Diffusion-OAMP (DDIM)Compression ratio (beta)=0.1, AWGN noise level (sigma)=0.52026.04 | 21.69 | — | 62.58 | |
| OAMP+BM3DCompression ratio (beta)=0.4, AWGN noise level (sigma)=0.52026.04 | 21.66 | — | 54.38 | |
| DDRMCompression ratio (beta)=0.1, AWGN noise level (sigma)=0.05, NFEs=502026.04 | 21.02 | — | 62.59 | |
| DDRMCompression ratio (beta)=0.1, AWGN noise level (sigma)=0.5, NFEs=502026.04 | 18 | — | 51.05 | |
| Diffusion-OAMP (FM)Compression ratio (beta)=0.1, AWGN noise level (sigma)=0.52026.04 | 17.7 | — | 46.63 | |
| OAMP+BM3DCompression ratio (beta)=0.1, AWGN noise level (sigma)=0.52026.04 | 17.58 | — | 37.83 | |
| DDRMCompression ratio (beta)=0.1, AWGN noise level (sigma)=0.05, NFEs=202026.04 | 17.35 | — | 50.45 | |
| DDRMCompression ratio (beta)=0.1, AWGN noise level (sigma)=0.5, NFEs=202026.04 | 15.48 | — | 43.67 | |
| GMA-VRFine-tuning Strategy=pre-trained2024.02 | — | -9.23 | — | |
| GMA-VRFine-tuning Strategy=Fine-tune encoder only2024.02 | — | -10.5 | — | |
| GMA-VRFine-tuning Strategy=Fine-tune encoder & decoder2024.02 | — | -17.3 | — | |
| GMA-VRFine-tuning Strategy=Knowledge Replay (ours)2024.02 | — | -14.4 | — | |
| MSH-VRFine-tuning Strategy=pre-trained2024.02 | — | 17.1 | — | |
| MSH-VRFine-tuning Strategy=Fine-tune encoder only2024.02 | — | 14.8 | — | |
| MSH-VRFine-tuning Strategy=Fine-tune encoder & decoder2024.02 | — | 8.95 | — | |
| MSH-VRFine-tuning Strategy=Knowledge Replay (ours)2024.02 | — | 9.57 | — | |
| Our modelFine-tuning Strategy=pre-trained2024.02 | — | -13.2 | — | |
| Our modelFine-tuning Strategy=Fine-tune encoder only2024.02 | — | -14.6 | — | |
| Our modelFine-tuning Strategy=Fine-tune encoder & decoder2024.02 | — | -19 | — | |
| Our modelFine-tuning Strategy=Knowledge Replay2024.02 | — | -16.6 | — |