Disentangled Feature Representation for Few-shot Image Classification

TL;DR

This paper introduces DFR, a novel framework for few-shot image classification that disentangles discriminative features from irrelevant variations, improving performance across multiple benchmarks including a new domain generalization dataset.

Contribution

The paper proposes a disentangled feature representation framework (DFR) that enhances few-shot learning by decoupling class-relevant features from irrelevant variations, and introduces a new dataset for domain generalization.

Findings

DFR achieves state-of-the-art results on all evaluated datasets.

Effective feature disentangling improves few-shot classification performance.

The framework is compatible with existing deep few-shot learning methods.

Abstract

Learning the generalizable feature representation is critical for few-shot image classification. While recent works exploited task-specific feature embedding using meta-tasks for few-shot learning, they are limited in many challenging tasks as being distracted by the excursive features such as the background, domain and style of the image samples. In this work, we propose a novel Disentangled Feature Representation framework, dubbed DFR, for few-shot learning applications. DFR can adaptively decouple the discriminative features that are modeled by the classification branch, from the class-irrelevant component of the variation branch. In general, most of the popular deep few-shot learning methods can be plugged in as the classification branch, thus DFR can boost their performance on various few-shot tasks. Furthermore, we propose a novel FS-DomainNet dataset based on DomainNet, for benchmarking the few-shot domain generalization tasks. We conducted extensive experiments to evaluate the proposed DFR on general and fine-grained few-shot classification, as well as few-shot domain generalization, using the corresponding four benchmarks, i.e., mini-ImageNet, tiered-ImageNet, CUB, as well as the proposed FS-DomainNet. Thanks to the effective feature disentangling, the DFR-based few-shot classifiers achieved the state-of-the-art results on all datasets.