Research map / Label-efficient and robust learning / Recognition and long-tail
Semantic Contrastive Bootstrapping for Single-Positive Multi-label Recognition
Cheng 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 |
|---|---|---|---|
| add recurrent semantic masked transformer with CAM guidance (ablation) | Scob · MS-COCO (1 positive) | mAP: 69.8 -> 73.8 | Improved |
| add CAM guidance inside SMT when contrastive learning with IPT is used (ablation) | Scob · MS-COCO (1 positive) | mAP: 73.9 -> 74.8 | Improved |
| add contrastive learning with random sampling to SMT with CAM (ablation) | Scob · MS-COCO (1 positive) | mAP: 73.8 -> 74.3 | Improved |
| add contrastive learning with IPT but without SMT (ablation) | Scob · MS-COCO (1 positive) | mAP: 69.8 -> 68.9 | Hurt |
| use instance priority tree instead of random negative sampling (ablation) | Scob · MS-COCO (1 positive) | mAP: 74.3 -> 74.8 | Improved |
| use larger CNN head with more parameters instead of SMT (ablation) | Large CNN · MS-COCO (1 positive) | mAP: 69.8 -> 70.9 | No clear effect |
| replace Scob contrastive learning with SoCo (ablation) | Baseline · MS-COCO (1 positive) | mAP: 69.8 -> 58.2 | Hurt |
| replace Scob contrastive learning with BYOL (ablation) | Baseline · MS-COCO (1 positive) | mAP: 69.8 -> 65.5 | Hurt |
| replace Scob with detection-based UP-DETR (ablation) | Baseline · MS-COCO (1 positive) | mAP: 69.8 -> 69.2 | No clear effect |
| optimize network and CAMs jointly instead of EM bootstrapping (ablation) | Scob · MS-COCO (1 positive) | mAP: 74.8 -> 69.7 | Hurt |
| train without CAM for 10 epochs then EM instead of EM from start (ablation) | Scob · MS-COCO (1 positive) | mAP: 74.8 -> 74.2 | Hurt |
| use ground truth masks instead of CAM guidance (ablation) | Scob · VOC 2012 | mAP: 89.8 -> 90.0 | No clear effect |
Closest papers by meaning
- Multi-Label Learning From Single Positive Labels (CVPR 2021 · similarity 0.87)
- Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels (AAAI 2022 · similarity 0.87)
- Structured Semantic Transfer for Multi-Label Recognition with Partial Labels (AAAI 2022 · similarity 0.87)
- Learning Semantic-Specific Graph Representation for Multi-Label Image Recognition (ICCV 2019 · similarity 0.87)
- Multi-Label Supervised Contrastive Learning (AAAI 2024 · similarity 0.87)
- L2B: Learning to Bootstrap Robust Models for Combating Label Noise (CVPR 2024 · 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.
