RAW Image Denoising on SID and ELD (test)
37.28PSNR (Sony)ADL
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| ADLIterations=300k, Optimizer=AdamW, Learning Rate=3e-3, Patch Size=256x2562024.11 | 37.28 | 87.1 | 37.58 | 87.2 | 36.74 | 86.6 | 36.45 | 86.8 | 37.01 | 86.8 | |
| Transfer learning2024.11 | 36.92 | 86.4 | 37.33 | 86.9 | 36.49 | 86.2 | 35.77 | 85.8 | 36.63 | 86.3 | |
| Blind2Unblind2024.11 | 36.71 | 86.6 | 36.57 | 86.6 | 35.88 | 85.7 | 35.49 | 85.5 | 36.16 | 86.1 | |
| MZSR2024.11 | 36.21 | 86.1 | 36.98 | 86.6 | 36.14 | 86 | 35.89 | 85.7 | 36.31 | 86.1 | |
| Prabhakar et al.2024.11 | 36.12 | 85.9 | 36.33 | 86.4 | 35.47 | 85.4 | 35.72 | 85.7 | 36.01 | 86.1 | |
| Fine-tuningTraining Protocol=training on source domain data and then fine-tuning on target domain data2024.11 | 35.94 | 85.7 | 36.37 | 86.2 | 35.22 | 85.3 | 35.63 | 85.5 | 35.79 | 85.7 | |
| BM3D2024.11 | 35.61 | 85.6 | 35.88 | 85.7 | 35.37 | 85.3 | 35.07 | 85.2 | 35.48 | 85.5 | |
| ZS-N2N2024.11 | 32.15 | 72.4 | 30.39 | 63.2 | 30.46 | 64.3 | 31.34 | 70.7 | 31.09 | 67.7 | |
| DIP2024.11 | 31.02 | 69.6 | 29.44 | 61.1 | 30.71 | 65.2 | 30.53 | 64.1 | 30.42 | 65 |