Task-Oriented Channel Attention for Fine-Grained Few-Shot Classification.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Fine-grained image classification is challenging due to shared appearances across classes.
• Identifying subtle, discriminative details is crucial but difficult with limited training data.

Purpose of the Study:

• To propose a novel attention method for fine-grained few-shot image classification.
• To enhance feature representation by focusing on class-discriminative and object-relevant details.

Main Methods:

• Task Discrepancy Maximization (TDM), a task-oriented channel attention method.
• Support Attention Module (SAM) and Query Attention Module (QAM) to highlight discriminative and relevant features.
• Instance Attention Module (IAM) for instance-wise feature highlighting in intermediate layers.

Main Results:

• TDM effectively produces task-adaptive features for accurate similarity measures.
• IAM complements TDM, improving performance in fine-grained few-shot tasks.
• IAM also demonstrates effectiveness in coarse-grained and cross-domain few-shot classifications.

Conclusions:

• The proposed TDM method, with SAM, QAM, and IAM, significantly advances fine-grained few-shot image classification.
• These attention mechanisms enable better utilization of limited data by focusing on critical visual details.