Rethinking Few-Shot Image Classification: a Good Embedding Is All You Need?

Abstract

The focus of recent meta-learning research has been on the development of learning algorithms that can quickly adapt to test time tasks with limited data and low computational cost. Few-shot learning is widely used as one of the standard benchmarks in meta-learning. In this work, we show that a simple baseline: learning a supervised or self-supervised representation on the meta-training set, followed by training a linear classifier on top of this representation, outperforms state-of-the-art few-shot learning methods. An additional boost can be achieved through the use of self-distillation. This demonstrates that using a good learned embedding model can be more effective than sophisticated meta-learning algorithms. We believe that our findings motivate a rethinking of few-shot image classification benchmarks and the associated role of meta-learning algorithms. Code is available at: http://github.com/WangYueFt/rfs/.

First two authors contributed equally. Project Page: https://people.csail.mit.edu/yuewang/projects/rfs/ Code: http://github.com/WangYueFt/rfs/

References in corpus (15)

• Distilling the Knowledge in a Neural Network
• Prototypical Networks for Few-shot Learning
• Momentum Contrast for Unsupervised Visual Representation Learning
• A Simple Neural Attentive Meta-Learner
• What Makes for Good Views for Contrastive Learning?
• Contrastive Multiview Coding
• Towards Understanding Knowledge Distillation
• Self-Distillation Amplifies Regularization in Hilbert Space
• Rapid Learning or Feature Reuse? Towards Understanding the Effectiveness of MAML
• Finding Task-Relevant Features for Few-Shot Learning by Category Traversal
• Infinite Mixture Prototypes for Few-Shot Learning
• Low-Shot Learning from Imaginary Data
• Contrastive Representation Distillation
• Adapted Deep Embeddings: A Synthesis of Methods for $k$-Shot Inductive Transfer Learning
• Few-Shot Learning with Global Class Representations

Cited by in corpus (28)

• Learning Transferable Visual Models From Natural Language Supervision
• Self-supervised Knowledge Distillation for Few-shot Learning
• FedBABU: Towards Enhanced Representation for Federated Image Classification
• The Lottery Tickets Hypothesis for Supervised and Self-supervised Pre-training in Computer Vision Models
• Self-Supervised Learning For Few-Shot Image Classification
• Independent Prototype Propagation for Zero-Shot Compositionality
• Bongard-LOGO: A New Benchmark for Human-Level Concept Learning and Reasoning
• Learning to Learn Variational Semantic Memory
• Dynamic Distillation Network for Cross-Domain Few-Shot Recognition with Unlabeled Data
• A Comprehensive Evaluation of Multi-task Learning and Multi-task Pre-training on EHR Time-series Data
• Dual-Awareness Attention for Few-Shot Object Detection
• Modeling and Optimization Trade-off in Meta-learning
• Meta-Learning with Adaptive Hyperparameters
• Machine learning with limited data
• S2SD: Simultaneous Similarity-based Self-Distillation for Deep Metric Learning
• Towards Fine-grained Visual Representations by Combining Contrastive Learning with Image Reconstruction and Attention-weighted Pooling
• A Foliated View of Transfer Learning
• Improving Few-Shot Learning with Auxiliary Self-Supervised Pretext Tasks
• Subspace Representation Learning for Few-shot Image Classification
• Model-Agnostic Graph Regularization for Few-Shot Learning
• Shot in the Dark: Few-Shot Learning with No Base-Class Labels
• Uncertainty-Aware Few-Shot Image Classification
• Improving the Generalization of Meta-learning on Unseen Domains via Adversarial Shift
• Enabling the Network to Surf the Internet
• Discrete Few-Shot Learning for Pan Privacy
• Improving Few-shot Learning with Weakly-supervised Object Localization
• Who calls the shots? Rethinking Few-Shot Learning for Audio
• Meta-Forecasting by combining Global Deep Representations with Local Adaptation