Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation

Abstract

Few-shot classification aims to recognize novel categories with only few labeled images in each class. Existing metric-based few-shot classification algorithms predict categories by comparing the feature embeddings of query images with those from a few labeled images (support examples) using a learned metric function. While promising performance has been demonstrated, these methods often fail to generalize to unseen domains due to large discrepancy of the feature distribution across domains. In this work, we address the problem of few-shot classification under domain shifts for metric-based methods. Our core idea is to use feature-wise transformation layers for augmenting the image features using affine transforms to simulate various feature distributions under different domains in the training stage. To capture variations of the feature distributions under different domains, we further apply a learning-to-learn approach to search for the hyper-parameters of the feature-wise transformation layers. We conduct extensive experiments and ablation studies under the domain generalization setting using five few-shot classification datasets: mini-ImageNet, CUB, Cars, Places, and Plantae. Experimental results demonstrate that the proposed feature-wise transformation layer is applicable to various metric-based models, and provides consistent improvements on the few-shot classification performance under domain shift.

arXiv commentICLR 2020 (Spotlight). Project page: http://vllab.ucmerced.edu/ym41608/projects/CrossDomainFewShot Code: https://github.com/hytseng0509/CrossDomainFewShot

References in corpus (16)

• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Prototypical Networks for Few-shot Learning
• Improved Regularization of Convolutional Neural Networks with Cutout
• The Effectiveness of Data Augmentation in Image Classification using Deep Learning
• A Learned Representation For Artistic Style
• AutoAugment: Learning Augmentation Policies from Data
• Domain Generalization via Invariant Feature Representation
• DropBlock: A regularization method for convolutional networks
• Smart Augmentation - Learning an Optimal Data Augmentation Strategy
• Few-Shot Learning with Graph Neural Networks
• Revisiting Batch Normalization For Practical Domain Adaptation
• Generalizing Across Domains via Cross-Gradient Training
• Few-Shot Learning with Metric-Agnostic Conditional Embeddings
• A Bayesian Data Augmentation Approach for Learning Deep Models
• Feature-Critic Networks for Heterogeneous Domain Generalization
• Dataset Augmentation in Feature Space

Cited by in corpus (41)

• BOIL: Towards Representation Change for Few-shot Learning
• Transductive Information Maximization For Few-Shot Learning
• Portrait Neural Radiance Fields from a Single Image
• MHFC: Multi-Head Feature Collaboration for Few-Shot Learning
• Cross-Domain Few-Shot Learning by Representation Fusion
• Explanation-Guided Training for Cross-Domain Few-Shot Classification
• Self-training for Few-shot Transfer Across Extreme Task Differences
• A Broader Study of Cross-Domain Few-Shot Learning
• Meta-Learning across Meta-Tasks for Few-Shot Learning
• Regularizing Generative Adversarial Networks under Limited Data
• StyleMeUp: Towards Style-Agnostic Sketch-Based Image Retrieval
• Few-shot Unsupervised Domain Adaptation with Image-to-class Sparse Similarity Encoding
• Learning a Universal Template for Few-shot Dataset Generalization
• Comparing Transfer and Meta Learning Approaches on a Unified Few-Shot Classification Benchmark
• Cross-Domain Few-Shot Classification via Adversarial Task Augmentation
• Regularizing Meta-Learning via Gradient Dropout
• Repurposing Pretrained Models for Robust Out-of-domain Few-Shot Learning
• EvoGrad: Efficient Gradient-Based Meta-Learning and Hyperparameter Optimization
• Large Margin Mechanism and Pseudo Query Set on Cross-Domain Few-Shot Learning
• Ranking Distance Calibration for Cross-Domain Few-Shot Learning
• DeltaGAN: Towards Diverse Few-shot Image Generation with Sample-Specific Delta
• SB-MTL: Score-based Meta Transfer-Learning for Cross-Domain Few-Shot Learning
• Learning to See Through Obstructions with Layered Decomposition
• Encoding Robustness to Image Style via Adversarial Feature Perturbations
• F2GAN: Fusing-and-Filling GAN for Few-shot Image Generation
• Online Hyperparameter Meta-Learning with Hypergradient Distillation
• A Primal-Dual Subgradient Approachfor Fair Meta Learning
• Streaming Self-Training via Domain-Agnostic Unlabeled Images
• Gaussian Process Meta Few-shot Classifier Learning via Linear Discriminant Laplace Approximation
• Improving the Generalization of Meta-learning on Unseen Domains via Adversarial Shift
• A Strong Baseline for Semi-Supervised Incremental Few-Shot Learning
• Combining Domain-Specific Meta-Learners in the Parameter Space for Cross-Domain Few-Shot Classification
• Modular Adaptation for Cross-Domain Few-Shot Learning
• Revisiting Few-shot Relation Classification: Evaluation Data and Classification Schemes
• Modeling Artistic Workflows for Image Generation and Editing
• Domain Conditional Predictors for Domain Adaptation
• MDFM: Multi-Decision Fusing Model for Few-Shot Learning
• Unlocking the Full Potential of Small Data with Diverse Supervision
• Self-Denoising Neural Networks for Few Shot Learning
• Detection of Adversarial Supports in Few-shot Classifiers Using Self-Similarity and Filtering
• Towards A Conceptually Simple Defensive Approach for Few-shot classifiers Against Adversarial Support Samples