AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Will Large-scale Generative Models Corrupt Future Datasets?

Ryuichiro Hataya, Han Bao, Hiromi Arai

ICCV 2023 · Image generation › Autoregressive and latent generation · approach: Generalizable fake image detection · 39 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace 80% of ImageNet training images with Stable Diffusion imagesResNet-50 · ImageNet valtop-1 accuracy: 75.7 -> 65.3Hurt
replace 20-40% of ImageNet images with generated imagesResNet-50 · ImageNet valtop-1 accuracy: 75.7 -> 74.5 (20%), 72.6 (40%)Hurt
use real WebVision images instead of generated images for contamination (ablation)ResNet-50 · ImageNet valtop-1 accuracy: IN/SD-80% 65.3 vs IN/WV-80% 68.3Improved
replace COCO captions images with generated imagesBLIP captioner · COCO testCIDEr: 1.335 -> 1.279 (80% generated)Hurt
replace ImageNet images with generated images in unconditional diffusion trainingIDDPM · ImageNet-64Recall@5: 0.644 -> 0.364 (100% generated)Hurt
replace ImageNet images with generated images in diffusion trainingIDDPM · ImageNet-64FID: 14.9 -> 11.0 (40%), 16.9 (100%)Mixed
use more complex, diverse prompts to generate SD-ImageNet (ablation)ResNet-50 · ImageNet valtop-1 accuracy: 15.7 -> 18.6No clear effect
contaminate with generated imagesResNet-50 · ImageNet-Raccuracy: 36.7 -> 30.0 (80% generated)Hurt
append generated images to full ImageNet instead of replacingResNet-50 · ImageNet valtop-1 accuracy: 75.7 -> 75.7 (+20%), 75.0 (+80%)No clear effect
fill missing 95% of ImageNet with generated images when only 5% real is availableResNet-50 · ImageNet valtop-1 accuracy: 44.7 -> 52.1Improved
fill missing COCO data with generated captions imagesBLIP captioner · COCO test (5% real)CIDEr: 1.290 -> 1.220Hurt
use self-supervised MAE pretraining instead of supervised training with generated dataViT-B · ImageNet val (linear probe)accuracy: SD-ImageNet 38.8 vs ImageNet 43.9Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy