Table 1 Quantitative comparison results on the SOTS outdoor datasets.

From: Online knowledge distillation network for single image dehazing

SOTS

Metrics

DCP

DCPDN

EPDN

AODNet

Grid

DA

MSBDN

PSD

Ours

Indoor

PSNR

19.95 dB

20.85 dB

25.09 dB

19.12 dB

32.16 dB

30.32 dB

32.67 dB

16.32 dB

30.92 dB

SSIM

0.872

0.875

0.932

0.845

0.984

0.981

0.983

0.729

0.988

LOE

227.94

331.51

258.30

266.54

188.55

247.55

168.34

320.59

131.10

NIQE

4.385

4.801

4.618

4.106

4.231

4.830

4.839

3.8310

3.319

Outdoor

PSNR

20.44 dB

20.08 dB

20.31 dB

21.47 dB

16.21 dB

22.59 dB

23.16 dB

15.15 dB

23.38 dB

SSIM

0.898

0.896

0.902

0.922

0.783

0.927

0.936

0.771

0.938

LOE

205.45

148.121

338.77

277.79

337.83

116.72

72.19

314.94

53.60

NIQE

3.1242

2.984

3.206

3.586

3.158

2.897

3.702

3.388

2.849

Model Parameters

0.94 M

35.86 M

0.09 M

0.96 M

54.59 M

28.71 M

6.2 M

2.58 M

  1. Number in bold, italics, and underline indicate the first, second, and third best results, respectively. DCP is a prior-based method without model parameters. Higher values of PSNR and SSIM; and lower values of LOE and NIQE represent better performance.