[AAAI'26] HABIT: Chrono-Synergia Robust Progressive Learning Framework for Composed Image Retrieval

Zixu Li1, Yupeng Hu1*, Zhiwei Chen1, Shiqi Zhang1, Qinlei Huang1, Zhiheng Fu1, Yinwei Wei1,
1School of Software, Shandong University
*Corresponding author.
Paper Complete source code

Abstract

MY ALT TEXT

Overcoming Unmentioned Visual Discrepancies via Chrono-Synergia

MY ALT TEXT

(a) presents an example of the CIR paradigm. (b) illustrates the commonly observed “unmentioned visual discrepancies” in CIR task, which increase the difficulty of identifying Noise Triplet Correspondence. (c) depicts our proposed Chrono-Synergia Mechanism.


Framework: cHrono-synergiA roBust progressIve learning framework for composed image reTrieval (HABIT)

MY ALT TEXT

HABIT consists of two modules: (a) Mutual Knowledge Estimation and (b) Dual-consistency Progressive Learning.

Experiment

MY ALT TEXT

Performance comparison on FashionIQ in terms of R@K (%). The best result under each noise ratio is highlighted in bold, while the second-best result is underlined.


MY ALT TEXT

Performance comparison on the CIRR test set in terms of R@K (%) and Rsub@K (%). The best and second-best results are highlighted in bold and underlined, respectively.

MY ALT TEXT

Ablation study on FashionIQ and CIRR datasets.


MY ALT TEXT

Case Study on CIRR and FashionIQ.

BibTeX


        @inproceedings{HABIT,
            title={HABIT: Chrono-Synergia Robust Progressive Learning Framework for Composed Image Retrieval},
            author={Li, Zixu and Hu, Yupeng and Chen, Zhiwei and Zhang, Shiqi and Huang, Qinlei and Fu, Zhiheng and Wei, Yinwei},
            booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
            year={2026}
        }