Learning to Compare: Relation Network for Few-Shot Learning

We present a conceptually simple, flexible, and general framework for few-shot learning, where a classifier must learn to recognise new classes given only few examples from each. Our method, called the Relation Network (RN), is trained end-to-end from scratch. During meta-learning, it learns to learn a deep distance metric to compare a small number of images within episodes, each of which is designed to simulate the few-shot setting. Once trained, a RN is able to classify images of new classes by computing relation scores between query images and the few examples of each new class without further updating the network. Besides providing improved performance on few-shot learning, our framework is easily extended to zero-shot learning. Extensive experiments on five benchmarks demonstrate that our simple approach provides a unified and effective approach for both of these two tasks.

PDF Abstract CVPR 2018 PDF CVPR 2018 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Few-Shot Image Classification CIFAR-FS 5-way (5-shot) Relation Networks* Accuracy 69.3 # 36
Few-Shot Image Classification CUB 200 5-way 1-shot Relation Net Accuracy 50.44 # 33
Few-Shot Image Classification CUB 200 5-way 5-shot Relation Net Accuracy 65.32 # 29
Few-Shot Image Classification Meta-Dataset Relation Networks Accuracy 53.315 # 20
Few-Shot Image Classification Meta-Dataset Rank Relation Networks Mean Rank 11.8 # 13
Few-Shot Image Classification Mini-Imagenet 10-way (1-shot) Relation Networks Accuracy 34.9 # 8
Few-Shot Image Classification Mini-Imagenet 10-way (5-shot) Relation Networks Accuracy 47.9 # 11
Few-Shot Image Classification Mini-Imagenet 5-way (1-shot) Relation Net (Sung et al., 2018) Accuracy 50.4 # 91
Few-Shot Image Classification OMNIGLOT - 1-Shot, 20-way Relation Net Accuracy 97.6% # 7
Few-Shot Image Classification OMNIGLOT - 1-Shot, 5-way Relation Net Accuracy 99.6 # 4
Few-Shot Image Classification OMNIGLOT - 5-Shot, 20-way Relation Net Accuracy 99.1% # 9
Few-Shot Image Classification OMNIGLOT - 5-Shot, 5-way Relation Net Accuracy 99.8 # 7
Few-Shot Image Classification Tiered ImageNet 10-way (1-shot) Relation Networks Accuracy 36.3 # 9
Few-Shot Image Classification Tiered ImageNet 10-way (5-shot) Relation Networks Accuracy 58.0 # 7

Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Few-Shot Image Classification Mini-ImageNet-CUB 5-way (1-shot) RelationNet (Sung et al., 2018) Accuracy 42.91 # 7
Image Classification Tiered ImageNet 5-way (5-shot) Relation Net Accuracy 71.31 # 2

Methods


No methods listed for this paper. Add relevant methods here