Research map / Label-efficient and robust learning / Recognition and long-tail
CDUL: CLIP-Driven Unsupervised Learning for Multi-Label Image Classification
Rabab Abdelfattah, Qing Guo, Xiaoguang Li, Xiaofeng Wang, Song Wang
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 local snippet alignment (3x3 crops) to CLIP global alignment for pseudo labels (ablation) | CLIP pseudo labels · VOC 2012 train | mAP: 85.3 -> 90.3 | Improved |
| add local snippet alignment to CLIP global alignment for pseudo labels (ablation) | CLIP pseudo labels · COCO train | mAP: 65.4 -> 72.8 | Improved |
| use min-max aggregator instead of average aggregator (ablation) | global-local pseudo labels · VOC 2012 train | mAP: 88.5 -> 90.3 | Improved |
| use min-max aggregator instead of max aggregator (ablation) | global-local pseudo labels · COCO train | mAP: 71.6 -> 72.8 | Improved |
| use larger CLIP backbone (ResNet-50x64) for pseudo labels (ablation) | CDUL · PASCAL VOC 2007 | mAP: 86.9 -> 89.0 | Improved |
| add Gaussian weighting module in gradient-alignment training (ablation) | CDUL · VOC 2012 and COCO | mAP: -0.5% VOC 2012, -0.4% COCO when removed | Improved |
| use soft pseudo labels instead of hard pseudo labels (ablation) | CDUL · VOC 2012 and COCO | mAP: -0.9% VOC 2012, -1.2% COCO with hard labels | Improved |
| train a ResNet-50 student on CLIP pseudo labels instead of using CLIP-GLA directly at inference (ablation) | CDUL · COCO | mAP: 63.6 -> 67.1 | Improved |
| train classifier with gradient-alignment on CLIP pseudo labels vs raw CLIP | CDUL · MS-COCO | mAP: 65.3 -> 69.2 | Improved |
Closest papers by meaning
- Classifier-guided CLIP Distillation for Unsupervised Multi-label Classification (CVPR 2025 · similarity 0.90)
- Unlocking the Power of Co-Occurrence in CLIP: A DualPrompt-Driven Method for Training-Free Zero-Shot Multi-Label Classification (ICLR 2026 · similarity 0.88)
- TagCLIP: A Local-to-Global Framework to Enhance Open-Vocabulary Multi-Label Classification of CLIP without Training (AAAI 2024 · similarity 0.88)
- Masked Unsupervised Self-training for Label-free Image Classification (ICLR 2023 · similarity 0.87)
- [CLS] is Not Enough: Multi-Label Recognition via Patch-Level Inference and Adaptive Aggregation (ICML 2026 · similarity 0.87)
- Recover and Match: Open-Vocabulary Multi-Label Recognition through Knowledge-Constrained Optimal Transport (CVPR 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.
