Table 2 Impact of reducing glyph instances on performance of the DPCDD Dataset

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

G of seen inpainting character

 

PSNR ↑ 

SSIM ↑ 

LPIPS ↓ 

FID ↓ 

StSC ↑ 

G = 5

CIDG4

18.7166

0.9289

0.0742

11.0120

0.9041

DE-GAN5

18.0239

0.7739

0.1572

95.0079

0.6941

CycleGAN2

13.9018

0.8078

0.1583

45.0151

0.7137

RubGAN3

14.1066

0.8342

0.1336

20.7753

0.7698

FD-Net25

18.4013

0.8448

0.1352

63.6833

0.8092

RCRN7

19.9264

0.9300

0.0647

8.9237

0.9218

Charformer6

19.3447

0.9267

0.0702

13.5724

0.9133

Uformer19

18.2975

0.9129

0.0885

16.1491

0.8560

Tformer17

19.0662

0.9020

0.0635

5.0784

0.9140

CENet29

21.4248

0.9050

0.0836

34.0152

0.9126

GSDM14

21.8589

0.9508

0.0421

2.9016

0.9560

CINet(ours)

22.5642

0.9501

0.0365

2.5185

0.9728

G = 3

CIDG4

18.6069

0.9300

0.0766

10.7084

0.8995

DE-GAN5

17.9095

0.7792

0.1579

100.5968

0.7162

CycleGAN2

13.6177

0.8071

0.1609

27.9348

0.6124

RubGAN3

14.2907

0.8260

0.1366

27.4403

0.7593

FD-Net25

17.7863

0.8308

0.1539

73.4424

0.7723

RCRN7

19.6830

0.9286

0.0671

9.5869

0.9224

Charformer6

19.6244

0.9184

0.0726

13.0601

0.9140

Uformer19

18.1853

0.9109

0.0906

16.3983

0.8451

Tformer17

18.7107

0.9197

0.0657

5.7012

0.9057

CENet29

21.1021

0.8967

0.0879

35.9235

0.8927

GSDM14

21.5331

0.9485

0.0442

3.1541

0.9488

CINet(ours)

22.1492

0.9525

0.0387

2.6915

0.9680

G = 1

CIDG4

18.4636

0.9276

0.0781

10.9620

0.9060

DE-GAN5

17.8112

0.7765

0.1623

96.9172

0.6861

CycleGAN2

11.4235

0.7148

0.2517

80.3395

0.5531

RubGAN3

13.3369

0.8087

0.1519

35.3300

0.7259

FD-Net25

14.4154

0.7813

0.2068

118.5517

0.5700

RCRN7

19.5665

0.9215

0.0704

11.5964

0.9156

Charformer6

16.5597

0.7348

0.1592

48.1134

0.7972

Uformer19

17.9328

0.9027

0.0957

19.7934

0.8398

Tformer17

18.8557

0.8502

0.0669

6.0376

0.9110

CENet29

21.3248

0.8957

0.0895

46.9806

0.8909

GSDM14

21.2570

0.9458

0.0464

3.3881

0.9472

CINet(ours)

21.7690

0.9504

0.0413

2.7523

0.9677

  1. The optimal results are shown in bold, and the sub-optimal results are underlined. G denotes glyph instance samples.