AI Paper Atlas

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

ReCON: Training-Free Acceleration for Text-to-Image Synthesis with Retrieval of Concept Prompt Trajectories

Chen-Yi Lu, Shubham Agarwal, Md Mehrab Tanjim, Kanak Mahadik, Anup Rao, Subrata Mitra, Shiv K Saini, Saurabh Bagchi, Somali Chaterji

ECCV 2024 · Image generation › Text-to-image diffusion · approach: Text-driven synthesis and semantic guidance · 0 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
retrieve and merge visual concept noises instead of a whole-prompt trajectoryretrieval-based SDXL acceleration · MS-COCO CLIP Score, NFE 35CLIP Score: 29.15 -> 30.19Improved
concept retrieval instead of text-based retrievalretrieval-based acceleration · Pick-a-PicCLIP Score: 27.67 -> 29.93Improved
noise-based ReDi retrieval compared with simple text-based retrieval on SDXLretrieval-based acceleration · MS-COCOCLIP Score: 29.15 -> 18.17Hurt
add visual-concept filtering (ablation)ReCon · Pick-a-PicCLIP Score: 29.06 -> 29.93Improved
add similarity-weighted aggregation instead of unweighted (ablation)ReCon · Pick-a-PicCLIP Score: 29.33 -> 29.93Improved
aggregate all concepts instead of only the Top-1 concept (ablation)ReCon · Pick-a-PicCLIP Score: 27.46 -> 29.93Improved
use ReCon across diffusion samplers (PNDM)ReCon · Pick-a-PicCLIP Score: 25.96 -> 27.68Improved
concept-based retrieval vs prompt retrieval at each text similarity interval (ablation)ReCon · text similarity 0.5-0.6, NFE 40PickScore: 20.38 -> 21.07Improved
concept retrieval at high text similarity (0.8-0.9) (ablation)ReCon · NFE 25PickScore: 20.43 -> 20.65Improved
ReCon vs Baseline on aestheticsReCon · MS-COCOAesthetics Score: 5.50 -> 5.52No clear effect
ReCon images in human evaluationReCon · user study, 200 participantsranked first: 75.5% ranked first, 16% second, 8.5% thirdImproved

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