/SuperResolution_Log

SuperResolution_Log

Training_logs

Training Setting:

  • batch size: 64
  • epochs: 1000
  • LR decay strategy: half at [250, 500, 750, 1000]
  • Loss: L1 loss
  • DIV2K
  • Data augmentation: Random rotation, random vertical flip, random horizontal flip

X4

Architecture # Parameters (K) MFLOPs Set5 (PSNR/SSIM) Set14 (PSNR/SSIM) B100 (PSNR/SSIM) Urban100 (PSNR/SSIM)
ECBSR 603 - 31.92 / 0.8946 28.34 / 0.7817 27.48 / 0.7393 25.81 / 0.7773
RLFN 543 - 32.24 / 0.8952 28.62 / 0.7813 27.60 / 0.7364 26.17 / 0.7877
07/03 - - 31.98 / 0.8917 28.45 / 0.7786 27.50 / 0.7338 25.77 / 0.7756

X4 from hang

Date Architecture # Parameters (K) MFLOPs Set5 (PSNR/SSIM) Set14 (PSNR/SSIM) B100 (PSNR/SSIM) Urban100 (PSNR/SSIM) Manga109 (PSNR/SSIM)
-- ECBSR 603 - 31.92 / 0.8946 28.34 / 0.7817 27.48 / 0.7393 25.81 / 0.7773
-- RLFN 543 - 32.24 / 0.8952 28.62 / 0.7813 27.60 / 0.7364 26.17 / 0.7877
-- RLFN we run - - 32.08 / - 28.26 / - 27.59 / - 26.12 / -
-- RLFN wo ESA - - 31.86 / - 28.19 / - 27.54 / - 25.85 / -
-- VDSR - - 19.13 / - 18.85 / - 20.40 / - 17.66 / -
-- VDSR w 4ESA - - 31.98 / - 28.19 / - 27.55 / - 25.96 / -
-- VDSR w 8ESA - - 32.05 / - 28.23 / - 27.58 / - 26.01 / -
7.7 VDSR w 8ESA res - - 32.06 / 0.8944 28.26 / 0.7823 27.59 / 0.7370 26.06 / 0.7863
7.7 VDSR w 8ESA res & c5 - - 32.13 / 0.8949 28.29 / 0.7828 27.60 / 0.7376 26.06 / 0.7867
7.7 RLFN inception - - 31.91 / 0.8914 28.11 / 0.7792 27.52 / 0.7345 25.88 / 0.7797
7.7 RLFN inception A - - 32.07 / 0.8943 28.28 / 0.7824 27.58 / 0.7371 26.05 / 0.7861
7.8 RLFN inception C (kill at 670 epoch) - - 32.03 / 0.8940 28.28 / 0.7828 27.57 / 0.7369 25.97 / 0.7840
7.8 RLFN inception A & cat ESA - - 32.08 / 0.8945 28.27 / 0.7826 27.60 / 0.7371 26.09 / 0.7867
7.8 RLFN inception A & cat inv - - 32.06 / 0.8944 28.28 / 0.7826 27.59 / 0.7371 26.08 / 0.7866
7.10 RLFN inception A & cat ESA with prelu 52 channel - - 32.13 / 0.8949 28.28 / 0.7831 27.60 / 0.7375 26.11 / 0.7879
7.10 RLFN inception A & cat ESA with 64 channel - - 32.17 / 0.8957 28.29 / 0.7834 27.64 / 0.7379 26.16 /0.7889
7.11 RLFN inception A & cat ESA 8 layer - - 32.20 / 0.8962 28.31 / 0.7838 27.62 / 0.7383 26.19 / 0.7901
7.11 RLFN inception A & cat ESA 10 layer - - 32.18 / 0.8960 28.33 / 0.7842 27.63 / 0.7388 26.27 / 0.7925
7.11 RLFN inception A & cat ESA 12 layer - - 32.22 / 0.8958 28.37 / 0.7841 27.64 / 0.7386 26.27 / 0.7920
7.11 RLFN inception A & cat ESA 16 layer - - 32.15 / 0.8954 28.32 / 0.7842 27.61 / 0.7383 26.14 / 0.7886
7.11 RLFN inception A & cat ESA cat attn 6 layer - - 32.26 / 0.8965 28.34 / 0.7840 27.64 / 0.7391 26.25 / 0.7911
-------- ----------------- ------------ ------ -------------------------- -------------------------- -------------------------- -------------------------- --------------------------
-- RLFN 543K - 32.24 / 0.8952 28.62 / 0.7813 27.60 / 0.7364 26.17 / 0.7877
-- RLFN we run 543K - 32.08 / - 28.26 / - 27.59 / - 26.12 / -
7.31 KernelFormer (919 epoch) Eval@Matlab 367K - 32.31 / 0.8964 28.67 / 0.7832 27.63 / 0.7380 26.22 / 0.7899 30.60 / 0.9100
7.29 ESRT 751K - 32.19 / 0.8947 28.69 / 0.7833 27.69 / 0.7379 26.39 / 0.7962 30.75 / 0.9100
7.31 KernelFormer (724 epoch) Eval@Matlab 671K - 32.35 / 0.8972 28.78 / 0.7858 27.68 / 0.7400 26.39 / 0.7958 30.86 / 0.9136
8.1 KernelFormer w Rep (506 epoch) Eval@Matlab 671K - 32.38 / 0.8974 28.76 / 0.7855 27.67 / 0.7398 26.41 / 0.7965 30.82 / 0.9131