Table 2.

Evaluations across benchmark data sets for efficient SR networks are presented. All results are computed on the Y-channel. The top and second-best performances (PSNR/dB↑&SSIM↑⁠) are denoted in red and blue colors, respectively

MethodsScale ×4Scale ×2
MURA-miniMURA-plusMURA-miniMURA-plus
Bicubic26.48/0.936226.80/0.931427.60/0.940527.66/0.9324
SRCNN (Dong et al., 2015) (TPAMI 2015)28.86/0.942628.99/0.937530.17/0.943330.48/0.9347
FSRCNN (Dong et al., 2016) (ECCV 2016)29.98/0.950830.22/0.943731.30/0.943431.97/0.9338
SRGAN (Ledig et al., 2017) (CVPR 2017)30.37/0.951130.68/0.944031.50/0.942332.17/0.9329
EDSR (Lim et al., 2017) (CVPRW 2017)30.29/0.950930.63/0.944031.51/0.943032.17/0.9339
RCAN (Zhang et al., 2018) (ECCV 2018)29.70/0.949329.89/0.942231.32/0.942532.01/0.9329
SwinIR (Liang et al., 2021) (ICCVW 2021)30.28/0.951430.60/0.944331.55/0.943332.17/0.9341
OverNet (Behjati et al., 2021) (WACV 2021)30.02/0.950330.22/0.943229.87/0.943931.82/0.9358
ShuffleMixer(base) (Sun et al., 2022) (NeurIPS 2022)29.88/0.949130.10/0.942531.35/0.942132.00/0.9327
ShuffleMixer(tiny) (Sun et al., 2022) (NeurIPS 2022)30.09/0.950630.35/0.943431.27/0.944231.84/0.9356
LBNet (Gao et al., 2022) (IJCAI 2022)30.34/0.950030.75/0.943131.50/0.940531.99/0.9310
DiVANet (Behjati et al., 2023) (PR 2023)30.39/0.951130.72/0.943931.32/0.941531.82/0.9320
Ours30.43/0.950930.82/0.943931.56/0.942232.18/0.9333
[ViewLarge]
This Feature Is Available To Subscribers Only

or Create an Account

Close Modal
Close Modal