Saccade and purify: Task adapted multi-view feature calibration network for few shot learning.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Few-shot image classification methods struggle with extracting complementary multi-view features and optimal feature selection.
• Existing approaches lack task-specific adaptation for feature calibration.

Purpose of the Study:

• To propose a novel Task-adapted Multi-view feature Calibration Network (TMCN) for improved few-shot image classification.
• To address the limitations in multi-view feature extraction and selection in current methods.

Main Methods:

• The TMCN simulates human visual saccade patterns for multi-view feature extraction, generating global, local grid, and random features.
• A global-local feature calibration module purifies features for stable non-local representations.
• A sampling feature fusion method and a multi-view feature calibrating module adaptively fuse features based on task information.

Main Results:

• The proposed TMCN achieves excellent performance on three public datasets.
• TMCN surpasses state-of-the-art methods in few-shot image classification tasks.
• The method demonstrates effective extraction and calibration of complementary multi-view features.

Conclusions:

• The TMCN offers a significant advancement in few-shot image classification by effectively leveraging multi-view features.
• Task-adapted feature calibration is crucial for enhancing classification performance.
• The approach provides a robust framework for future research in adaptive feature learning.