Table 5.
The accuracy of classification and detection tasks connected with demosaicking methods. The ‘Origin’ item indicates the original accuracy of the pre-trained model MobileNet v1 and SSD300. The bolded two methods in each section are those that achieve better performance among all algorithms.
| Algorithm | MobileNet v1 | SSD300 | |
|---|---|---|---|
| Top1 (%) | Top5 (%) | mAP (%) | |
| Origin | 71.11 | 89.84 | 75.77 |
| AHD [48] | 64.79 | 85.67 | 75.41 |
| DLMMSE [49] | 64.06 | 85.44 | 75.14 |
| RI [5] | 64.25 | 85.65 | 75.16 |
| MLRI [6] | 64.36 | 85.70 | 75.21 |
| ARI [7] | 64.40 | 85.74 | 75.06 |
| Tan [11] | 65.02 | 86.04 | 75.59 |
| Kokkinos [13] | 64.43 | 85.76 | 75.56 |
| Cui [14] | 64.50 | 85.80 | 75.49 |
| Ours (L1) | 64.11 | 85.49 | 75.16 |
| Ours (L2) | 64.43 | 85.78 | 75.22 |
| Ours (L3) | 64.56 | 85.83 | 75.44 |