Fan Liu, Xiaoying Ji, Chuanyi Zhang, Xinlei Zhang, Liying Zhang, Xiaohan Yu · ACM International Conference on Multimedia (ACM MM) 2026 · 2026
DOI: 10.1145/3767308.3835288
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Cross-modal retrieval aims to establish semantic alignments across different modalities. However, existing methods frequently assume the availability of perfectly paired training data. This assumption ignores the inevitable noisy correspondence (NC) within real-world datasets, which degrades the performance of retrieval. To address these limitations, we propose the Multi-phase Collaborative Representation (MCR) framework. To the best of our knowledge, MCR is the first framework that introduces cross-modal collaborative representation to tackle the NC problem, accurately identifying true correspondences. Specifically, MCR reconstructs a query feature using a linear combination of training samples within the batch from the other modality. By optimizing the objective of collaborative representation, we leverage the reconstruction coefficients to measure the semantic correlation between each pair. Furthermore, we propose a multi-phase iterative process that enhances the feature representations of clean samples and dynamically filters noisy ones. This process produces reliable matching confidence weights to effectively quantify the noise. Extensive experiments on three widely-used benchmark datasets, including Flickr30K, MS-COCO, and Conceptual Captions, validate the effectiveness of our proposed approach. The code is available at https://github.com/Ji-Xiaoying/MM26_MCR.
ACM International Conference on Multimedia (ACM MM) 2026 · 0 citations
ACM International Conference on Multimedia (ACM MM) 2026 · 0 citations
No comments yet — start the discussion below.