@inproceedings{du-etal-2026-openidc,
title = "{O}pen{IDC}: Reconstructing the Evidence Chain for Lightweight Image Difference Captioning",
author = "Du, Jiabao and
Hou, Saihui and
Liu, Yuan and
Wang, Jiawei and
Huang, Yongzhen",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1209/",
pages = "25516--25529",
ISBN = "979-8-89176-462-0",
abstract = "Image Difference Captioning (IDC) has produced many task-specific models, but their gains are often entangled with isolated implementations, data processing, training protocols, and evaluation settings. As a result, the field accumulates method-level scores more easily than reusable structural evidence. This paper proposes OpenIDC, a flexible and extensible codebase that reconstructs a unified evidence chain for lightweight task-specific IDC. OpenIDC standardizes data pipelines, model interfaces, training strategies, and evaluation protocols, enabling 8 representative method families with 9 executable variants across 3 benchmarks to be reproduced, compared, decomposed, and diagnosed under consistent conditions. Beyond reproduction, OpenIDC advances IDC from final-score comparison to normalized design-space diagnosis. Based on this diagnosis, we introduce SIDFormer, a compact Spatially-guided Invariant-Difference Transformer that serves as a structural answer to lightweight IDC by organizing spatial anchoring, pseudo-change suppression, and reliable difference evidence through functional fusion. Experiments show improved reproducibility and comparability, together with competitive SIDFormer performance across the evaluated benchmarks. Code, reproduced methods, and models will be released."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="du-etal-2026-openidc">
<titleInfo>
<title>OpenIDC: Reconstructing the Evidence Chain for Lightweight Image Difference Captioning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiabao</namePart>
<namePart type="family">Du</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Saihui</namePart>
<namePart type="family">Hou</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuan</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiawei</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yongzhen</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Image Difference Captioning (IDC) has produced many task-specific models, but their gains are often entangled with isolated implementations, data processing, training protocols, and evaluation settings. As a result, the field accumulates method-level scores more easily than reusable structural evidence. This paper proposes OpenIDC, a flexible and extensible codebase that reconstructs a unified evidence chain for lightweight task-specific IDC. OpenIDC standardizes data pipelines, model interfaces, training strategies, and evaluation protocols, enabling 8 representative method families with 9 executable variants across 3 benchmarks to be reproduced, compared, decomposed, and diagnosed under consistent conditions. Beyond reproduction, OpenIDC advances IDC from final-score comparison to normalized design-space diagnosis. Based on this diagnosis, we introduce SIDFormer, a compact Spatially-guided Invariant-Difference Transformer that serves as a structural answer to lightweight IDC by organizing spatial anchoring, pseudo-change suppression, and reliable difference evidence through functional fusion. Experiments show improved reproducibility and comparability, together with competitive SIDFormer performance across the evaluated benchmarks. Code, reproduced methods, and models will be released.</abstract>
<identifier type="citekey">du-etal-2026-openidc</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1209/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>25516</start>
<end>25529</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T OpenIDC: Reconstructing the Evidence Chain for Lightweight Image Difference Captioning
%A Du, Jiabao
%A Hou, Saihui
%A Liu, Yuan
%A Wang, Jiawei
%A Huang, Yongzhen
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F du-etal-2026-openidc
%X Image Difference Captioning (IDC) has produced many task-specific models, but their gains are often entangled with isolated implementations, data processing, training protocols, and evaluation settings. As a result, the field accumulates method-level scores more easily than reusable structural evidence. This paper proposes OpenIDC, a flexible and extensible codebase that reconstructs a unified evidence chain for lightweight task-specific IDC. OpenIDC standardizes data pipelines, model interfaces, training strategies, and evaluation protocols, enabling 8 representative method families with 9 executable variants across 3 benchmarks to be reproduced, compared, decomposed, and diagnosed under consistent conditions. Beyond reproduction, OpenIDC advances IDC from final-score comparison to normalized design-space diagnosis. Based on this diagnosis, we introduce SIDFormer, a compact Spatially-guided Invariant-Difference Transformer that serves as a structural answer to lightweight IDC by organizing spatial anchoring, pseudo-change suppression, and reliable difference evidence through functional fusion. Experiments show improved reproducibility and comparability, together with competitive SIDFormer performance across the evaluated benchmarks. Code, reproduced methods, and models will be released.
%U https://aclanthology.org/2026.findings-emnlp.1209/
%P 25516-25529
Markdown (Informal)
[OpenIDC: Reconstructing the Evidence Chain for Lightweight Image Difference Captioning](https://aclanthology.org/2026.findings-emnlp.1209/) (Du et al., Findings 2026)
ACL