Research map / Vision-language and multimodal / Cross-modal learning
BiCro: Noisy Correspondence Rectification for Multi-Modality Data via Bi-Directional Cross-Modal Similarity Consistency
Shuo Yang, Zhaopan Xu, Kai Wang, Yang You, Hongxun Yao, Tongliang Liu, Min Xu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use BiCro soft correspondence labels instead of binary labels (ablation) | SGRAF · Flickr30K 40% noise | Image-to-text R@1: 72.0 -> 74.6 | Improved |
| Use BiCro soft correspondence labels instead of binary labels (ablation) | SGRAF · Flickr30K 40% noise | Text-to-image R@1: 53.3 -> 55.5 | Improved |
| Replace GMM with BMM for modeling per-sample losses (ablation) | BiCro · Flickr30K 40% noise | Image-to-text R@1: 72.0 -> 74.6 | Improved |
| Remove co-teaching (ablation) | BiCro · Flickr30K 40% noise | Image-to-text R@1: 74.6 -> 69.9 | Hurt |
| Remove warmup (ablation) | BiCro · Flickr30K 40% noise | Image-to-text R@1: 74.6 -> 54.3 | Hurt |
| Set soft correspondence labels below a mismatch threshold to zero | BiCro · Flickr30K 40% noise | Sum | Improved |
| Use BiCro rather than DECL | SGRAF · Flickr30K 60% noise | Sum: 450.6 -> 465.7 | Improved |
| Use BiCro rather than DECL | SGRAF · MS-COCO 1K 60% noise | Sum: 502.5 -> 505.8 | Improved |
| Use BiCro rather than DECL | SGRAF · CC152K | Sum: 364.3 -> 368.9 | Improved |
Later papers that compared with this method
4 later accepted papers report BiCro as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Rethinking Cross-Modal Anchor Alignment for Mitigating Error Accumulation (CVPR 2026 · better on 1 of 1 shared result)
- Robust Noisy Correspondence Learning with Equivariant Similarity Consistency (CVPR 2024 · better on 3 of 4 shared results)
- Mitigating Noisy Correspondence by Geometrical Structure Consistency Learning (CVPR 2024 · better on 1 of 1 shared result)
- Cross-modal Active Complementary Learning with Self-refining Correspondence (NeurIPS 2023 · better on 3 of 3 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
- Noisy Correspondence Rectification via Asymmetric Similarity Learning (AAAI 2025 · similarity 0.91)
- Robust Noisy Correspondence Learning with Equivariant Similarity Consistency (CVPR 2024 · similarity 0.90)
- AlignMamba: Enhancing Multimodal Mamba with Local and Global Cross-modal Alignment (CVPR 2025 · similarity 0.88)
- On the Value of Cross-Modal Misalignment in Multimodal Representation Learning (NeurIPS 2025 · similarity 0.88)
- SoftCLIP: Softer Cross-Modal Alignment Makes CLIP Stronger (AAAI 2024 · similarity 0.87)
- Learning to Rematch Mismatched Pairs for Robust Cross-Modal Retrieval (CVPR 2024 · similarity 0.87)
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.
