sRGB Noise Modeling on SIDD (test)
-3.502NLLNoise Flow
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Noise FlowData Requirement=Clean/noisy image pairs2022.06 | -3.502 | 0.0267 | |
| Noise2NoiseFlowTraining=Joint noise model training, Data Requirement=Noisy image pairs2022.06 | -3.501 | 0.0265 | |
| N2N+NFTraining=Separately trained Noise2Noise denoiser and NoiseFlow, Data Requirement=Noisy image pairs2022.06 | -3.459 | 0.0363 | |
| Cam. NLFDescription=Camera noise level function2022.06 | -3.282 | 0.0578 | |
| AWGNDescription=Additive White Gaussian Noise2022.06 | -2.874 | 0.4815 | |
| Our modelNumber of parameters=61602022.06 | 3.072 | 0.044 | |
| Noise Flow-LargeNumber of parameters=66182022.06 | 3.288 | 0.227 | |
| Noise flowNumber of parameters=23302022.06 | 3.311 | 0.198 | |
| Full GaussianNumber of parameters=5252022.06 | 3.608 | 0.085 | |
| Heteroscedastic (NLF)Number of parameters=722022.06 | 3.642 | 0.088 | |
| Diagonal GaussianNumber of parameters=1502022.06 | 3.678 | 0.079 | |
| Isotropic GaussianNumber of parameters=502022.06 | 3.703 | 0.091 |