Table 2 Ablation Study of the REPLKA Module: The impact of different configurations, including the absence of the REPLKA module, single-branch module, two-branch module, three-branch module, and four-branch module, on \(\times\)2 Super-Resolution tasks performed by REPLKASR is investigated. The best metrics are highlighted in bold.
From: Reparameterizable large kernel attention networks for infrared image super-resolution
REPLKA Module | Scale | Params (K) | Multi- Adds (G) | Set5 PSNR/SSIM | Set14 PSNR/SSIM | BSD100 PSNR/SSIM | Urban100 PSNR/SSIM | Manga109 PSNR/SSIM |
|---|---|---|---|---|---|---|---|---|
No REPLKA | \(\times\)2 | 111 | 6.4 | 31.60/0.8867 | 28.27/0.7738 | 27.34/0.7279 | 25.38/0.7613 | 29.47/0.8943 |
Single- Branch | 113 | 6.6 | 31.63/0.8873 | 28.29/0.7745 | 27.35/0.7284 | 25.45/0.7641 | 29.57/0.8957 | |
Two- Branch | 113 | 6.6 | 31.64/0.8877 | 28.30/0.7747 | 27.34/0.7286 | 25.45/0.7643 | 29.57/0.8959 | |
Three- Branch | 113 | 6.6 | 31.66/0.8877 | 28.30/0.7749 | 27.36/0.7288 | 25.45/0.7645 | 29.60/0.8958 | |
Four- Branch | 113 | 6.6 | 31.66/0.8882 | 28.31/0.7749 | 27.36/0.7290 | 25.46/0.7647 | 29.62/0.8960 |
