AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Understanding and Mitigating Copying in Diffusion Models

Gowthami Somepalli, Vasu Singla, Micah Goldblum, Jonas Geiping, Tom Goldstein

NeurIPS 2023 · Image generation › Text-to-image diffusion · approach: Inference-time T2I improvement

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
Multiple captions (20 BLIP captions per image) at trainingSD 2.1 finetuned · LAION-10k no duplicationdataset similarity (lower better): 0.638 -> 0.420Improved
Multiple captions at trainingSD 2.1 finetuned · LAION-10k image + caption duplicationdataset similarity (lower better): 0.663 -> 0.470*Improved
Random token replacement at inferenceStable Diffusion · LAION-10k image + caption duplicationdataset similarity (lower better): 0.663 -> 0.555Improved
Random number addition at inferenceStable Diffusion · LAION-10k image duplicationdataset similarity (lower better): 0.576 -> 0.497Improved
Random token replacement during trainingSD 2.1 finetuned · LAION-10k image duplicationdataset similarity (lower better): 0.576 -> 0.645Hurt
Random token replacement during trainingSD 2.1 finetuned · LAION-10k no duplicationdataset similarity (lower better): 0.638 -> 0.638No clear effect
Gaussian noise on text embeddings during trainingSD 2.1 finetuned · LAION-10k no duplicationFID (lower better): 18.72 -> 20.63Hurt
Multiple captions during trainingSD 2.1 finetuned · LAION-10k no duplicationFID (lower better): 18.72 -> 16.63Improved
Random token addition at inferenceStable Diffusion · LAION-10k no duplicationCLIP score: 30.52 -> 29.54Hurt
Use diverse (BLIP) captions instead of a fixed caption (ablation)SD 2.1 finetuned on Imagenette · ImagenetteFID (lower better): 40.6 -> 17.74Improved
Partial duplication (resampled captions per duplicate) instead of full duplication, rising ddf (ablation)SD 2.1 finetuned · LAION-10k / Imagenettedataset similarityImproved
Train the text encoder during finetuning (ablation)SD 2.1 finetuned · Imagenette / LAION-10kdataset similarityHurt

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