Research map / Label-efficient and robust learning / Recognition and long-tail
PatchCT: Aligning Patch Set and Label Set with Conditional Transport for Multi-Label Image Classification
Miaoge Li, Dongsheng Wang, Xinyang Liu, Zequn Zeng, Ruiying Lu, Bo Chen, Mingyuan Zhou
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 layer-wise conditional transport loss between patch set and label set (ablation) | PatchCT · MS-COCO | mAP: 84.7 -> 88.3 | Improved |
| add layer-wise conditional transport loss (ablation) | PatchCT · Pascal VOC 2007 test | mAP: 94.9 -> 97.1 | Improved |
| add layer-wise conditional transport loss (ablation) | PatchCT · NUS-WIDE | mAP: 65.8 -> 68.1 | Improved |
| use the prompt 'A photo of {label}' instead of the bare label name (ablation) | PatchCT · MS-COCO | mAP: 88.0 -> 88.3 | Improved |
| use the prompt template instead of label name only (ablation) | PatchCT · Pascal VOC 2007 test | mAP: 96.7 -> 97.1 | Improved |
| vary top-k patch count from 50 to 250 (ablation) | PatchCT · Pascal VOC 2007 test | mAP: 96.8 to 97.2 | No clear effect |
| start the CT alignment at an early layer (l_s < 6) rather than late | PatchCT · MS-COCO and Pascal VOC 2007 | mAP | Improved |
| start the CT alignment from layer 1 | PatchCT · MS-COCO and Pascal VOC 2007 | mAP | Hurt |
| replace cross-attention alignment with CT alignment (PatchCT vs M3TR) | multi-label classifier · MS-COCO | mAP: 87.5 -> 88.3 | Improved |
| PatchCT vs MlTr-1 on NUS-WIDE overall F1 | PatchCT · NUS-WIDE (ALL) | OF1: 75.8 -> 74.7 | Mixed |
Closest papers by meaning
- [CLS] is Not Enough: Multi-Label Recognition via Patch-Level Inference and Adaptive Aggregation (ICML 2026 · similarity 0.89)
- Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels (AAAI 2022 · similarity 0.88)
- Cross-Modality Attention with Semantic Graph Embedding for Multi-Label Classification (AAAI 2020 · similarity 0.88)
- Counterfactual Reasoning for Multi-Label Image Classification via Patching-Based Training (ICML 2024 · similarity 0.86)
- Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning (CVPR 2025 · similarity 0.86)
- Incomplete Multi-View Multi-label Learning via Disentangled Representation and Label Semantic Embedding (CVPR 2025 · similarity 0.85)
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.
