Research map / Label-efficient and robust learning / Noisy labels
HaNa: Hardness and Noise-Aware Robust Cross-modal Retrieval
Fangming Zhong, Haiquan Yu, Cun Zhu, Suhua Zhang
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 weighted contrastive loss (vs standard contrastive loss) (ablation) | HaNa · Flickr30K 40% noise | image-to-text R@1: 76.0 -> 88.5 | Improved |
| add weighted contrastive loss (ablation) | HaNa · Flickr30K 40% noise | text-to-image R@1: 57.5 -> 75.1 | Improved |
| add momentum update of weights (ablation) | HaNa · Flickr30K 40% noise | image-to-text R@1: 87.6 -> 88.5 | Improved |
| add momentum update of weights (ablation) | HaNa · Flickr30K 40% noise | text-to-image R@1: 74.1 -> 75.1 | Improved |
| add hardness weights on clean samples (vs weight 1) (ablation) | HaNa · Flickr30K 40% noise | image-to-text R@1: 86.5 -> 88.5 | Improved |
| add hardness weights on clean samples (ablation) | HaNa · Flickr30K 40% noise | text-to-image R@1: 73.5 -> 75.1 | Improved |
| add ANRL (ablation) | HaNa · Flickr30K 40% noise | image-to-text R@1: 87.3 -> 88.5 | Improved |
| add ANRL (ablation) | HaNa · Flickr30K 40% noise | text-to-image R@1: 74.3 -> 75.1 | Improved |
| HaNa versus NPC at 60% noise | HaNa · Flickr30K 60% noise | image-to-text R@1: 83.0 -> 87.1 | Improved |
| HaNa versus NPC on real-world noise | HaNa · CC120K | RSUM: 517.6 -> 521.5 | Improved |
| increase noise ratio from 20% to 60% for HaNa | HaNa · Flickr30K | image-to-text R@1: 89.0 -> 87.1 | Mixed |
| HaNa versus NPC on MS-COCO text-to-image R@1 at 20% noise | HaNa · MS-COCO 1K, 20% noise | text-to-image R@1: 66.3 -> 69.0 | Improved |
Closest papers by meaning
- Negative Pre-aware for Noisy Cross-Modal Matching (AAAI 2024 · similarity 0.89)
- Learning with Noisy Correspondence for Cross-modal Matching (NeurIPS 2021 · similarity 0.89)
- Noisy Correspondence Learning with Self-Reinforcing Errors Mitigation (AAAI 2024 · similarity 0.88)
- Learning to Rematch Mismatched Pairs for Robust Cross-Modal Retrieval (CVPR 2024 · similarity 0.88)
- Noisy Correspondence Rectification via Asymmetric Similarity Learning (AAAI 2025 · similarity 0.87)
- Robust Self-Paced Hashing for Cross-Modal Retrieval with Noisy Labels (AAAI 2025 · similarity 0.86)
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.
