Table 6 Comparison of performance on real-inscription images.

From: A structural information-guided cross-modal method for damaged inscription inpainting via vision-language models

DID

PSNR ↑ 

SSIM ↑ 

LPIPS ↓ 

FID ↓ 

StSc ↑ 

CIDG4

19.7501

0.9407

0.0628

25.5439

0.9098

DE-GAN5

19.9100

0.9076

0.1430

122.9757

0.8952

CycleGAN2

16.9535

0.8826

0.1056

43.6596

0.6900

RubGAN3

17.6579

0.5975

0.1841

115.0252

0.8180

FD-Net25

18.9690

0.8811

0.1743

147.9551

0.8355

RCRN7

18.7303

0.9123

0.1003

78.1304

0.8559

Charformer6

18.1647

0.4742

0.2935

204.4901

0.7948

Uformer19

18.5368

0.9312

0.0585

18.5603

0.8355

Tformer17

20.0371

0.9449

0.0475

15.8084

0.9229

CENet29

20.4213

0.8961

0.1285

92.8850

0.9112

GSDM14

20.7975

0.9347

0.0917

81.6815

0.9199

CINet

21.7686

0.9564

0.0370

16.2541

0.9534

  1. The optimal results are shown in bold, and the sub-optimal results are underlined.