Research map / Vision-language and multimodal / Cross-modal learning
Noisy Correspondence Learning with Modality Gap Direction Correction
Wuyuqing Wang, Zeyuan Gu, Erkun Yang
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 |
|---|---|---|---|
| add warmup training before noisy-correspondence training (ablation) | MGCS · Flickr30K 20% noise | image-to-caption R@1: 7.9 -> 82.2 | Improved |
| add SADR similarity rectification (ablation) | MGCS · Flickr30K 20% noise | image-to-caption R@1: 79.8 -> 82.2 | Improved |
| add SADR similarity rectification (ablation) | MGCS · Flickr30K 20% noise | caption-to-image R@1: 59.5 -> 62.0 | Improved |
| add dynamic regularization (ablation) | MGCS · Flickr30K 20% noise | image-to-caption R@1: 81.6 -> 82.2 | Improved |
| add dynamic regularization (ablation) | MGCS · Flickr30K 20% noise | caption-to-image R@1: 60.7 -> 62.0 | Improved |
| use MGCS instead of best baseline on real-noise CC152K | MGCS · CC152K | R@ sum: 379.1 -> 385.9 | Improved |
| use MGCS instead of best baseline on Flickr30K 20% noise | MGCS · Flickr30K 20% noise | R@ sum: 508.1 -> 515.4 | Improved |
| use MGCS instead of best baseline on MS-COCO 60% noise | MGCS · MS-COCO 60% noise | R@ sum: 515.0 -> 526.9 | Improved |
| use MGCS instead of best baseline on Flickr30K 60% noise, caption-to-image R@1 | MGCS · Flickr30K 60% noise | caption-to-image R@1: MGCS 55.3 vs SREM 54.0; but DisNCL-like baselines lower; sum 490.3 vs 480.3 | Mixed |
| add SAD computation to training | MGCS · Flickr30K training cost | seconds per iteration: 4.6 -> 6.8 | Hurt |
Closest papers by meaning
- Robust Noisy Correspondence Learning with Equivariant Similarity Consistency (CVPR 2024 · similarity 0.92)
- Noisy Correspondence Rectification via Asymmetric Similarity Learning (AAAI 2025 · similarity 0.91)
- Noisy Correspondence Learning with Self-Reinforcing Errors Mitigation (AAAI 2024 · similarity 0.90)
- Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion (ICLR 2025 · similarity 0.90)
- Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMs (NeurIPS 2026 · similarity 0.89)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.89)
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.
