AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Seeing Through Words: Controlling Visual Retrieval Quality with Language

Jianglin Lu, Simon Jenni, Kushal Kafle, Jing Shi, Handong Zhao, Yun Fu

ICLR 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Composed and text-based image retrieval

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
condition LLM query completion on quality levels instead of random scores (ablation)QCQC · Flickr2.4M, CoCa, Aes High/Rel HighAve Rel: 0.365 -> 0.390Improved
complete short queries with fine-tuned quality-conditioned LLM instead of prefix onlyQCQC · Flickr2.4M, CoCa, Aes High/Rel HighAve Aes: 4.735 -> 5.270Improved
complete short queries with QCQC instead of prefix onlyQCQC · MS-COCO, Aes High/Rel HighAve Aes: 4.817 -> 5.024Improved
use pretrained LLM for completion without finetuning (ablation)PT baseline · Flickr2.4M and MS-COCOAve Aes / Ave RelHurt
fine-tune LLM on image descriptions (random quality conditions) (ablation)FT baseline · Flickr2.4M, CoCaAve Rel: 0.350 -> 0.365Improved
raise quality condition from Low to High (ablation)QCQC · Flickr2.4M, CoCa, Rel Low vs Rel HighAve Rel: 0.355 -> 0.391Improved
raise aesthetic condition from Low to High (ablation)QCQC · Flickr2.4M, CoCa, Aes L vs HAve Aes: 4.458 -> 5.222Improved
use QCQC instead of GPT-4o completionQCQC · Flickr2.4M, CoCa, Aes High/Rel HighAve Rel: 0.361 -> 0.390Improved
use QCQC instead of GPT-4o completion on MS-COCOQCQC · MS-COCO, Aes High/Rel HighAve Aes: 5.048 -> 5.024No clear effect
use QCQC instead of post-retrieval aesthetic rerankingQCQC · Flickr2.4M, k=10Ave Rel: 0.341 -> 0.366Improved
increase rerank candidate pool k from 1 to 10 (ablation)Rerank baseline · Flickr2.4MAve Aes / Ave Rel: Aes 4.735 -> 5.313, Rel 0.350 -> 0.341Mixed
use five quality levels instead of three (ablation)QCQC · Flickr2.4M, CoCaAve AesImproved

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