AI Paper Atlas

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

ELITE: Encoding Visual Concepts into Textual Embeddings for Customized Text-to-Image Generation

Yuxiang Wei, Yabo Zhang, Zhilong Ji, Jinfeng Bai, Lei Zhang, Wangmeng Zuo

ICCV 2023 · Image generation › Text-to-image diffusion · approach: Personalized and region-controlled generation · 391 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
learn multiple words from multiple CLIP layers instead of from the single deepest layer, and use primary word only (ablation)ELITE global mapping · DreamBenchCLIP-T: 0.212 -> 0.257Improved
learn multiple words from the deepest single layer rather than one word (ablation)ELITE global mapping · DreamBenchDINO-I: 0.546 -> 0.431 (full embeddings)Hurt
add local mapping network (ablation)ELITE · DreamBenchCLIP-I / DINO-I: CLIP-I 0.699 -> 0.762; DINO-I 0.486 -> 0.652Improved
add local mapping network (ablation)ELITE · DreamBenchCLIP-T: 0.257 -> 0.255No clear effect
increase fusion weight lambda of local attention from 0 to 1.2 (ablation)ELITE local mapping · DreamBenchCLIP-I, DINO-I, CLIP-TMixed
place deepest layer word first versus middle or last among layer words (ablation)ELITE global mapping · DreamBenchCLIP-T: 0.249 / 0.247 -> 0.257Improved
add local attention map reweighting by primary word attention (ablation)ELITE local mapping · qualitativeeditabilityImproved
add global mapping network to local-only mapping (ablation)ELITE · qualitativeconcept fidelityImproved
replace per-concept optimization with feed-forward encodercustomized text-to-image generation · DreamBenchencoding time: 6 min -> 0.05s (vs Custom Diffusion)Improved
use ELITE instead of Custom Diffusion / DreamBoothcustomized generation · DreamBenchCLIP-I / DINO-I: CLIP-I 0.801 -> 0.762; DINO-I 0.695 -> 0.652Hurt

Later papers that compared with this method

15 later accepted papers report ELITE as a baseline in their results tables; 15 of them report a better number on most of the results they share with it.

7 more not listed.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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