Table 1.
Quantitative comparison of foreground and background image restoration performance.
Model | Params (M) | FLOPs (G) | Foreground | Background | ||
---|---|---|---|---|---|---|
PSNR | SSIM | PSNR | SSIM | |||
DnCNN | 0.671 | 11.01 | 35.81 | 0.973 | 35.92 | 0.975 |
RRDB | 16.624 | 252.00 | 35.93 | 0.974 | 36.12 | 0.978 |
DPIR | 32.640 | 35.893 | 35.98 | 0.976 | 36.29 | 0.979 |
SwinIR | 11.504 | 197.00 | 36.06 | 0.977 | 36.31 | 0.981 |
Restormer | 26.112 | 38.721 | 36.13 | 0.978 | 36.32 | 0.981 |
STUNet | 17.79 | 34.27 | 36.21 | 0.978 | 36.31 | 0.982 |
TORNet (Ours) | 17.342 | 30.777 | 36.38 | 0.979 | 36.43 | 0.982 |
Bold values indicate the best performance for each column.