Table 4 Performance comparison of lightweight SR methods on the RK3588NPU. Tested on public datasets. Inference time is measured based on the output image size of 1280\(\times\)720. The best and second-best performances are highlighted in Italic and bold, respectively.
From: Reparameterizable large kernel attention networks for infrared image super-resolution
REPLKA Module | Scale | M3FD-15 PSNR/SSIM | Iray-15 PSNR/SSIM | Iray-boat PSNR/SSIM | Iray-traffic PSNR/SSIM | time consuming(ms) |
|---|---|---|---|---|---|---|
ESPCN-uint8 | \(\times\)4 | 24.76/0.6552 | 27.56/0.7543 | 30.83/0.9031 | 31.50/0.9125 | 44.24 |
FSRCNN-uint8 | 24.76/0.6616 | 27.74/0.7761 | 31.11/0.9138 | 31.81/0.9214 | 53.32 | |
IMDN-RTC-uint8 | 24.78/0.6620 | 27.76/0.7732 | 31.30/0.9164 | 32.00/0.9237 | 51.67 | |
ECBSR-uint8 | 24.81/0.6656 | 27.74/0.7759 | 31.48/0.9187 | 32.28/0.9262 | 37.22 | |
REPLKASR-uint8 | 24.82/0.6659 | 27.78/0.7784 | 31.49/0.9189 | 32.33/0.9266 | 37.59 |
