Image Manipulation research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

LSUN Bedroom VINE-watermarked

Asyrp

100Accuracy

May 20, 2026

LSUN Church VINE-watermarked

Asyrp

1Accuracy

May 20, 2026

AFHQ-Dog HiDDeN-watermarked (test)

Asyrp

100Accuracy

May 20, 2026

CelebA HiDDeN-watermarked (test)

Asyrp

100Accuracy

May 20, 2026

Image manipulation Few-shot (In Distribution)

InstaManip

19.81CLIP-Dir

Feb 26, 2026

Few-shot image manipulation (Out of Distribution)

InstaManip

18.27CLIP Directional Score

Feb 26, 2026

Image Manipulation Evaluation Dataset 135 images

DALL-E 2

78.47CLIPScore

Feb 26, 2026

User Study

SEGA

80Multi-Conditioning

Feb 26, 2026

General Face Image Editing

TediGAN

20Inference Time

Feb 26, 2026

LSUN Church 256 (test)

DiffusionCLIP

0.2Sdir

Feb 26, 2026

CelebA-HQ 256 (test)

DiffusionCLIP

0.17Sdir

Feb 26, 2026

CelebA-HQ General cases, All domains (test)

DiffusionCLIP

62.47Preference Rate vs StyleGAN-NADA (pSp)

Feb 26, 2026

CelebA-HQ General cases Out-of-domain (test)

DiffusionCLIP

71.03Preference Rate (vs StyleGAN-NADA)

Feb 26, 2026

CelebA-HQ General cases In-domain (test)

DiffusionCLIP

58.05Preference Rate vs StyleGAN-NADA (+ Restyle w pSp)

Feb 26, 2026

CelebA-HQ Hard cases All domains (test)

DiffusionCLIP

73.1Preference Rate (vs StyleGAN-NADA)

Feb 26, 2026

CelebA-HQ Hard cases, Out-of-domain (test)

DiffusionCLIP

79.6Preference Rate vs StyleGAN-NADA (pSp)

Feb 26, 2026

CelebA-HQ Hard cases, In-domain (test)

DiffusionCLIP

69.85Preference Rate vs StyleGAN-NADA (pSp)

Feb 26, 2026