Bi-Directional Ensemble Feature Reconstruction Network for Few-Shot Fine-Grained Classification.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Fine-grained few-shot image classification requires learning discriminative features from limited data.
• Conventional few-shot learning methods often increase intra-class variations, hindering fine-grained classification.
• Existing reconstruction-based methods primarily address inter-class variations, neglecting intra-class variations.

Purpose of the Study:

• To develop a method that simultaneously addresses inter-class and intra-class variations in fine-grained few-shot image classification.
• To enhance the learning of subtle and discriminative features crucial for fine-grained tasks.
• To improve the performance of few-shot image classification models.

Main Methods:

• Introduction of a bi-reconstruction mechanism: reconstructing the query set from the support set (increasing inter-class variations) and the support set from the query set (reducing intra-class variations).
• Integration of a self-reconstruction module to further enhance feature discriminability.
• Application of the snapshot ensemble method within the episodic learning strategy to boost performance without additional training costs.

Main Results:

• The proposed bi-reconstruction mechanism effectively accommodates both inter-class and intra-class variations.
• The self-reconstruction module further refines feature discriminability.
• Consistent and considerable performance improvements were observed across general, cross-domain, and fine-grained few-shot image classification datasets.

Conclusions:

• The bi-reconstruction mechanism is a significant advancement for fine-grained few-shot image classification.
• The method effectively learns more subtle and discriminative features, outperforming existing approaches.
• The proposed techniques offer a robust solution for challenging few-shot image classification scenarios.