Table 3 Ablation study on the Thangka dataset, evaluating the effectiveness of the proposed MACCA and CDSA modules
From: MACColor: multi-scale and cross-dimensional attention for thangka image colorization
Method | 256 × 256 | 512 × 512 | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
FID↓ | SSIM↑ | CF↑ | ΔCF↓ | PSNR↑ | FID↓ | SSIM↑ | CF↑ | ΔCF↓ | PSNR↑ | |
Baseline | 17.7934 | 0.8597 | 48.2252 | 5.6100 | 21.7974 | 18.6270 | 0.8555 | 49.0850 | 6.9384 | 22.5366 |
MACCA replaced by CBAM | 17.6924 | 0.8673 | 47.8573 | 1.8254 | 21.3842 | 18.5365 | 0.8466 | 48.3675 | 2.5282 | 22.5362 |
CDSA replaced by CoTNet | 17.9582 | 0.8463 | 48.1204 | 1.5623 | 21.9608 | 18.5340 | 0.8325 | 49.1754 | 1.7203 | 22.2546 |
w/o MACCA | 17.4352 | 0.8657 | 49.5783 | 3.4563 | 23.3563 | 18.0253 | 0.8634 | 49.1528 | 3.5243 | 22.8572 |
w/o CDSA | 17.4537 | 0.8735 | 49.2564 | 3.6351 | 22.5463 | 18.4936 | 0.8462 | 50.6342 | 4.5486 | 22.6345 |
Full (MACCA+CDSA) | 17.0173 | 0.8925 | 50.7837 | 2.5010 | 24.0659 | 17.9443 | 0.8709 | 51.4480 | 2.5523 | 23.8310 |
