AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

iSEARLE: Improving Textual Inversion for Zero-Shot Composed Image Retrieval

Lorenzo Agnolucci, Alberto Baldrati, Alberto Del Bimbo, Marco Bertini

TPAMI 2025 · Vision-language and multimodal › CLIP and zero-shot · approach: Attribute-based zero-shot learning · 47 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
distill OTI into feed-forward network phi (iSEARLE vs iSEARLE-OTI)iSEARLE · FashionIQ val (average, B/32)Recall@10: 25.06 -> 24.40No clear effect
improved version over conference SEARLESEARLE · CIRCO test (B/32)mAP@10: 9.94 -> 11.24Improved
improved version over conference SEARLESEARLE · CIRR test (B/32)Recall@1: 24.00 -> 25.23Improved
improved version over conference SEARLE-XLSEARLE-XL · FashionIQ val (average, L/14)Recall@10: 25.56 -> 27.52Improved
improved version over conference SEARLE-XLSEARLE-XL · Domain conversion (average, L/14)Recall@10: 12.04 -> 15.52Improved
train phi with FashionIQ training set as pre-training datasetiSEARLE-XL · FashionIQ val (average)Recall@10 and Recall@50: 29.07 and 49.67Improved
use distilled phi instead of OTI for L/14iSEARLE-XL · COCO object compositionRecall@1: 17.54 -> 15.01Hurt
scale ViT-B/32 to ViT-L/14iSEARLE · CIRR testRecall@1: 25.23 -> 25.28No clear effect
add reference image to caption (Image + Text vs Text-only)CLIP baseline · CIRR test (B/32)Recall@1: 21.81 -> 11.71Hurt
add reference image to caption (Image + Text vs Text-only)CLIP baseline · CIRCO test (B/32)mAP@10: 2.67 -> 3.25Improved
use pseudo-word token instead of generated captioniSEARLE · CIRCO test (B/32)mAP@10: 5.77 -> 11.24Improved

Later papers that compared with this method

1 later accepted paper reports iSEARLE as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

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