Table 3 Quantitatively compare with the state-of-the-art methods in the image super-resolution field on the Infrared Images benchmark dataset. The best and second-best performances are highlighted in Italic and bold, respectively.
From: Reparameterizable large kernel attention networks for infrared image super-resolution
Method | Scale | Params (K) | Multi- Adds (G) | FLOPs (G) | M3FD-15 PSNR/SSIM | Iray-15 PSNR/SSIM | Iray-boat PSNR/SSIM | Iray-traffic PSNR/SSIM |
|---|---|---|---|---|---|---|---|---|
ESPCN | \(\times\)2 | 21 | 2.28 | 4.55 | 30.44/0.8820 | 33.86/0.9123 | 30.83/0.9031 | 31.50/0.9125 |
FSRCNN | 12 | 3 | 6 | 30.75/0.8879 | 34.21/0.9200 | 31.11/0.9138 | 31.81/0.9214 | |
IMDN-RTC | 19 | 2.29 | 4.57 | 30.92/0.8920 | 34.45/0.9226 | 31.30/0.9164 | 32.00/0.9237 | |
ECBSR | 94 | 10.9 | 21.8 | 31.29/0.8956 | 34.65/0.9244 | 31.48/0.9187 | 32.28/0.9262 | |
MAN-tiny | 150 | 4.2 | 8.4 | 31.17/0.9018 | 34.61/0.9272 | 31.35/0.9157 | 31.98/0.9219 | |
SMFANet | 186 | 20.5 | 41 | 31.13/0.9016 | 34.55/0.9268 | 31.34/0.9157 | 31.93/0.9218 | |
REPLKASR | 105 | 21.2 | 42.4 | 31.31/0.8964 | 34.73/0.9247 | 31.49/0.9189 | 32.33/0.9266 | |
ESPCN | \(\times\)4 | 24 | 0.72 | 1.44 | 25.06/0.6906 | 27.90/0.7921 | 25.76/0.7542 | 26.43/0.7889 |
FSRCNN | 12 | 2.3 | 4.6 | 25.13/0.6954 | 28.01/0.7974 | 25.79/0.7589 | 26.48/0.7928 | |
IMDN-RTC | 21 | 0.61 | 1.22 | 25.12/0.6990 | 28.05/0.8013 | 25.81/0.7611 | 26.47/0.7947 | |
ECBSR | 98 | 2.83 | 5.65 | 25.14/0.7000 | 28.11/0.8038 | 25.84/0.7628 | 26.49/0.7964 | |
MAN-tiny | 150 | 4.2 | 8.4 | 25.16/0.6945 | 28.16/0.7976 | 25.88/0.7497 | 26.52/0.7802 | |
SMFANet | 197 | 5.5 | 11 | 25.17/0.6955 | 28.19/0.7990 | 25.87/0.7495 | 26.52/0.7808 | |
REPLKASR | 113 | 6.6 | 13.2 | 25.17/0.7010 | 28.17/0.8045 | 25.91/0.7641 | 26.53/0.7968 |
