Research map / Label-efficient and robust learning / Recognition and long-tail
Label2Label: A Language Modeling Framework for Multi-Attribute Learning
Wanhua Li, Zhexuan Cao, Jianjiang Feng, Jie Zhou, Jiwen Lu
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 Transformer-decoder attribute query network instead of FC head (ablation) | FC Head · LFWA (ResNet-50) | error (%): 13.63 -> 13.36 | Improved |
| add image-conditioned masked language model on top of AQN (ablation) | AQN · LFWA (ResNet-50) | error (%): 13.36 -> 12.49 | Improved |
| add IC-MLM on top of AQN (ablation) | AQN · LFWA (ViT-B) | error (%): 13.32 -> 12.79 | Improved |
| add IC-MLM on top of AQN (ablation) | AQN · PA100K | mA (%): 80.89 -> 82.24 | Improved |
| add IC-MLM on top of AQN (ablation) | AQN · Clothing Attributes Dataset | total accuracy (%): 92.29 -> 92.87 | Improved |
| add plain masked language model (no image condition) after AQN (ablation) | AQN · LFWA (ResNet-50) | error (%): 13.36 -> 13.32 | No clear effect |
| add image condition to MLM (two-stage, fixed features) (ablation) | MLM · LFWA (ResNet-50) | error (%): 13.32 -> 13.01 | Improved |
| add joint training of IC-MLM with attribute query network (ablation) | IC-MLM · LFWA (ResNet-50) | error (%): 13.01 -> 12.49 | Improved |
| add random masking with ratio 0.1 (ablation) | Label2Label · LFWA (ResNet-50) | error (%): 12.55 -> 12.49 | Improved |
| use attribute-specific [mask] embedding instead of zero vector (ablation) | Label2Label · LFWA (ResNet-50) | error (%): 12.60 -> 12.49 | Improved |
| add position embeddings to word tokens (ablation) | Label2Label · LFWA (ResNet-50) | error (%): 12.49 -> 12.51 | No clear effect |
| add 2D position embeddings to visual features (ablation) | Label2Label · LFWA (ResNet-50) | error (%): 12.98 -> 12.49 | Improved |
Later papers that compared with this method
1 later accepted paper reports Label2Label as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Selective and Orthogonal Feature Activation for Pedestrian Attribute Recognition (AAAI 2024 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Understanding Partial Multi-Label Learning via Mutual Information (NeurIPS 2021 · similarity 0.86)
- Navigating Real-World Partial Label Learning: Unveiling Fine-Grained Images with Attributes (AAAI 2024 · similarity 0.84)
- In-Context Learning Learns Label Relationships but Is Not Conventional Learning (ICLR 2024 · similarity 0.83)
- Attribute-formed Class-specific Concept Space: Endowing Language Bottleneck Model with Better Interpretability and Scalability (CVPR 2025 · similarity 0.83)
- Black Sheep in the Herd: Playing with Spuriously Correlated Attributes for Vision-Language Recognition (ICLR 2025 · similarity 0.83)
- Adaptive Integration of Partial Label Learning and Negative Learning for Enhanced Noisy Label Learning (AAAI 2024 · similarity 0.83)
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.
