Table 4 Impact of reducing glyph instances and style samples on CINet performance.

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

G of seen

inpainting character

S of seen

inpainting character

PSNR ↑ 

SSIM ↑ 

LPIPS ↓ 

FID ↓ 

StSc ↑ 

5

2629

22.5642

0.9501

0.0365

2.5185

0.9728

1315

21.5947

0.9493

0.0434

3.0026

0.9694

657

21.0008

0.9282

0.0482

3.7203

0.9638

329

20.3057

0.9244

0.0538

4.2230

0.9483

3

2629

22.1492

0.9525

0.0387

2.6915

0.9680

1315

21.1227

0.9474

0.0461

3.3452

0.9634

657

20.3279

0.9416

0.0533

4.2516

0.9567

329

19.5161

0.9350

0.0610

5.5980

0.9405

1

2629

21.7690

0.9504

0.0413

2.7523

0.9677

1315

20.4333

0.9418

0.0511

4.1197

0.9568

657

20.2268

0.9414

0.0547

4.5300

0.9535

329

19.3160

0.9155

0.0646

6.0300

0.9437