Research map / Vision-language and multimodal / Human and embodied modeling
Adaptive Offline Quintuplet Loss for Image-Text Matching
Tianlang Chen, Jiajun Deng, Jiebo Luo
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 |
|---|---|---|---|
| train with offline hard negatives and adaptive offline quintuplet loss | BFAN · MS-COCO 5K test | sentence retrieval R@1: 52.9 -> 57.3 | Improved |
| train with AdapOffQuin | VSRN · MS-COCO 1K test | rsum: 516.8 -> 521.4 | Improved |
| train with AdapOffQuin | SCAN · MS-COCO 1K test | rsum: 507.9 -> 511.2 | Improved |
| train with AdapOffQuin | BFAN · Flickr30K 1K test | rsum: 470.4 -> 486.7 | Improved |
| add offline quintuplet (derived non-shared-anchor negative pairs) to offline triplet loss (ablation) | BFAN · MS-COCO 1K test | rsum: 513.4 -> 514.7 | Improved |
| add adaptive hierarchical penalization to offline quintuplet loss (ablation) | BFAN · MS-COCO 1K test | rsum: 514.7 -> 517.2 | Improved |
| add offline triplet loss to online triplet loss (ablation) | VSRN · MS-COCO 1K test | rsum: 516.8 -> 518.6 | Improved |
| keep online loss term alongside offline negatives instead of only offline (ablation) | VSRN · MS-COCO 1K test | sentence retrieval R@1: 0.7 -> 77.5 | Improved |
| retrain from scratch in second round instead of fine-tuning (ablation) | VSRN · MS-COCO 1K test | sentence retrieval R@1: 74.5 -> 77.5 | Improved |
| derive extra negative pairs from offline instead of online negatives (ablation) | BFAN · MS-COCO 1K test | sentence retrieval R@1: 75.3 -> 77.3 | Improved |
| add offline-negative scores into online term for hierarchy (ablation) | VSRN · MS-COCO 1K test | sentence retrieval R@1: 77.1 -> 77.5 | Improved |
| vary top list size h from 200 to 1000 (ablation) | VSRN · MS-COCO 1K test | sentence retrieval R@1: 77.1 -> 77.3 | No clear effect |
Closest papers by meaning
- Expanding the Scope of Negatives: Boosting Image-Text Matching with Negatives Distribution Guided Learning (AAAI 2025 · similarity 0.92)
- Noisy Correspondence Rectification via Asymmetric Similarity Learning (AAAI 2025 · similarity 0.86)
- Robust Noisy Correspondence Learning with Equivariant Similarity Consistency (CVPR 2024 · similarity 0.85)
- Achieving Ensemble-Like Performance in a Single Model: A Feature Diversification Framework for Image-Text Matching (AAAI 2025 · similarity 0.85)
- Aligning Composed Query with Image via Discriminative Perception from Negative Correspondences (AAAI 2025 · similarity 0.84)
- On the Generalization of Handwritten Text Recognition Models (CVPR 2025 · similarity 0.84)
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.
