AI Paper Atlas

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

VSC: Visual Search Compositional Text-to-Image Diffusion Model

Do Huu Dat, Nam Hyeon-Woo, Po-Yuan Mao, Tae-Hyun Oh

ICCV 2025 · Image generation › Text-to-image diffusion · approach: Personalized and region-controlled generation · 3 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
add VSC visual prototype fusionSD3.5 · T2I-CompBenchBLIP-VQA harmonic mean: 0.639 -> 0.727Improved
add VSC visual prototype fusionSD2.1 · T2I-CompBenchBLIP-VQA harmonic mean: 0.467 -> 0.608Improved
add VSC visual prototype fusionSD1.4 · T2I-CompBenchBLIP-VQA harmonic mean: 0.375 -> 0.568Improved
use VSC instead of SynGen with five binding pairsSD2.1 · Prompts with 5 binding pairsBLIP-VQA score: 0.1189 -> 0.1901Improved
use VSC instead of SynGen with four binding pairsSD3.5 · Prompts with 4 binding pairsBLIP-VQA score: 0.5664 -> 0.6183Improved
use VSC instead of SynGen with three binding pairsSD3.5 · Prompts with 3 binding pairsBLIP-VQA score: 0.8081 -> 0.8132No clear effect
train only on color category instead of all categories (ablation)VSC SD3.5 · T2I-CompBench colorBLIP-VQA score: 0.85 -> 0.89Improved
train only on color category instead of all categories (ablation)VSC SD3.5 · T2I-CompBench textureBLIP-VQA score: 0.79 -> 0.74Hurt
train only on color category (transfer to shape) versus untrained baseline (ablation)VSC SD2.1 · T2I-CompBench shapeBLIP-VQA score: 0.42 -> 0.46Improved
scale synthetic training set from 5k to 30k images (ablation)VSC SD2.1 and SD3.5 · T2I-CompBenchBLIP-VQA score: 5,000 to 30,000 imagesImproved
scale synthetic training set beyond 30k to 90k images (ablation)VSC SD2.1 and SD3.5 · T2I-CompBenchBLIP-VQA score: 30,000 to 90,000 imagesNo clear effect
train last 2-3 layers of image encoder versus frozen encoder (ablation)VSC · Localization losslocalization loss: converges to approximately -0.04Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

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