Intro
Preface
Organization
Contents - Part V
Recognition: Feature Detection, Indexing, Matching, and Shape Representation
Improving Few-shot Learning by Spatially-aware Matching and CrossTransformer*-12pt
1 Introduction
2 Related Work
3 Approach
3.1 Spatially-aware Few-shot Learning
3.2 Self-supervised Scale and Scale Discrepancy
3.3 Transformer-Based Spatially-Aware Pipeline
4 Experiments
4.1 Datasets
4.2 Performance Analysis
5 Conclusions
References
AONet: Attentional Occlusion-Aware Network for Occluded Person Re-identification*-12pt

1 Introduction
2 Related Works
3 Attentional Occlusion-Aware Network
3.1 Landmark Patterns and Memorized Dictionary
3.2 Attentional Latent Landmarks
3.3 Referenced Response Map
3.4 Occlusion Awareness
3.5 Training and Inference
4 Experiments
4.1 Datasets and Implementations
4.2 Comparisons to State-of-the-Arts
4.3 Ablation Studies
5 Conclusion
References
FFD Augmentor: Towards Few-Shot Oracle Character Recognition from Scratch
1 Introduction
2 Related Works
2.1 Oracle Character Recognition
2.2 Few-Shot Learning

2.3 Data Augmentation Approaches
2.4 Non-Rigid Transformation
3 Methodology
3.1 Problem Formulation
3.2 Overview of Framework
3.3 FFD Augmentor
3.4 Training with FFD Augmentor
4 Experiments
4.1 Experimental Settings
4.2 Evaluation of FFD Augmented Training
4.3 Further Analysis of FFD Augmentor
4.4 Applicability to Other Problems
5 Conclusion
References
Few-shot Metric Learning: Online Adaptation of Embedding for Retrieval*-12pt
1 Introduction
2 Related Work
2.1 Metric Learning
2.2 Few-shot Classification
3 Few-shot Metric Learning

3.1 Metric Learning Revisited
3.2 Problem Formulation of Few-shot Metric Learning
4 Methods
4.1 Simple Fine-Tuning (SFT)
4.2 Model-Agnostic Meta-Learning (MAML)
4.3 Meta-Transfer Learning (MTL)
4.4 Channel-Rectifier Meta-Learning (CRML)
5 Experiments
5.1 Experimental Settings
5.2 Effectiveness of Few-shot Metric Learning
5.3 Influence of Domain Gap Between Source and Target
5.4 Few-shot Metric Learning vs. Few-shot Classification
5.5 Results on miniDeepFashion
6 Conclusion
References

3D Shape Temporal Aggregation for Video-Based Clothing-Change Person Re-identification*-12pt
1 Introduction
2 Related Work
3 Method
3.1 Parametric 3D Human Estimation
3.2 Identity-Aware 3D Shape Generation
3.3 Difference-Aware Shape Aggregation
3.4 Appearance and Shape Fusion
4 VCCR Dataset
4.1 Collection and Labelling
4.2 Statistics and Comparison
4.3 Protocol
5 Experiments
5.1 Implementation Details
5.2 Evaluation on CC Re-Id Datasets
5.3 Evaluation on Short-Term Re-Id Datasets
5.4 Ablation Study
6 Conclusion
References