Research map / Label-efficient and robust learning / Recognition and long-tail
Adapting Dense Vision-Language Relationships for Multi-Label Classification With Partial Label
C Q Chen, Yifan Zhao, Jia Li
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 |
|---|---|---|---|
| replace attention pooling with dense features (resolution 14) (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 67.3 -> 69.8 | Improved |
| add densely contrastive adaptor to dense features (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 69.8 -> 73.2 | Improved |
| add language guidance (interactive decoder) to dense features (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 69.8 -> 77.0 | Improved |
| add class-specific prompt tuning on top of language guidance (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 77.0 -> 77.1 | No clear effect |
| add densely contrastive adaptor on top of language guidance and prompt (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 77.1 -> 79.1 | Improved |
| freeze CLIP encoder and add adaptor instead of naive fine-tuning (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 34.0 -> 79.1 | Improved |
| use MoCo-style contrastive learning (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 67.3 -> 75.0 | Improved |
| use BYOL-style contrastive learning (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 67.3 -> 71.2 | Improved |
| use DetCo-style contrastive learning (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 67.3 -> 59.2 | Hurt |
| use ORL-style contrastive learning (ablation) | LDSA · MS-COCO (ablation, EMA 1.0) | mAP: 67.3 -> 39.7 | Hurt |
| add EMA to DualCoOp baseline (ablation) | DualCoOp · MS-COCO (partial labels, avg over 10-90%) | mAP: 81.9 -> 78.5 | Hurt |
Closest papers by meaning
- Partial Multi-View Multi-Label Classification via Semantic Invariance Learning and Prototype Modeling (ICML 2024 · similarity 0.88)
- Language-Driven Cross-Modal Classifier for Zero-Shot Multi-Label Image Recognition (ICML 2024 · similarity 0.87)
- Classifier-guided CLIP Distillation for Unsupervised Multi-label Classification (CVPR 2025 · similarity 0.87)
- Recover and Match: Open-Vocabulary Multi-Label Recognition through Knowledge-Constrained Optimal Transport (CVPR 2025 · similarity 0.87)
- Efficient and Context-Aware Label Propagation for Zero-/Few-Shot Training-Free Adaptation of Vision-Language Model (ICLR 2025 · similarity 0.87)
- Incomplete Multi-View Multi-label Learning via Disentangled Representation and Label Semantic Embedding (CVPR 2025 · 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.
