Leveraging the Feature Distribution in Transfer-based Few-Shot Learning

6 Jun 2020  ·  Yuqing Hu, Vincent Gripon, Stéphane Pateux ·

Few-shot classification is a challenging problem due to the uncertainty caused by using few labelled samples. In the past few years, many methods have been proposed to solve few-shot classification, among which transfer-based methods have proved to achieve the best performance. Following this vein, in this paper we propose a novel transfer-based method that builds on two successive steps: 1) preprocessing the feature vectors so that they become closer to Gaussian-like distributions, and 2) leveraging this preprocessing using an optimal-transport inspired algorithm (in the case of transductive settings). Using standardized vision benchmarks, we prove the ability of the proposed methodology to achieve state-of-the-art accuracy with various datasets, backbone architectures and few-shot settings.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Few-Shot Image Classification CIFAR-FS 5-way (1-shot) PT+MAP Accuracy 87.69 # 5
Few-Shot Image Classification CIFAR-FS 5-way (5-shot) PT+MAP Accuracy 90.68 # 7
Few-Shot Image Classification CUB 200 5-way 1-shot PT+MAP Accuracy 91.55% # 7
Few-Shot Image Classification CUB 200 5-way 5-shot PT+MAP Accuracy 93.99 # 6
Few-Shot Image Classification Dirichlet CUB-200 (5-way, 1-shot) PT-MAP 1:1 Accuracy 65.1 # 8
Few-Shot Image Classification Dirichlet CUB-200 (5-way, 5-shot) PT-MAP 1:1 Accuracy 71.3 # 8
Few-Shot Image Classification Dirichlet Mini-Imagenet (5-way, 1-shot) PT-MAP 1:1 Accuracy 60.6 # 6
Few-Shot Image Classification Dirichlet Mini-Imagenet (5-way, 5-shot) PT-MAP 1:1 Accuracy 67.1 # 10
Few-Shot Image Classification Dirichlet Tiered-Imagenet (5-way, 1-shot) PT-MAP 1:1 Accuracy 64.1 # 8
Few-Shot Image Classification Dirichlet Tiered-Imagenet (5-way, 5-shot) PT-MAP 1:1 Accuracy 70.0 # 9
Few-Shot Image Classification Mini-ImageNet - 1-Shot Learning PT+MAP Accuracy 82.92% # 1
Few-Shot Image Classification Mini-Imagenet 5-way (10-shot) PT+MAP Accuracy 90.03 # 1
Few-Shot Image Classification Mini-Imagenet 5-way (1-shot) PT+MAP (transductive) Accuracy 82.92 # 10
Few-Shot Image Classification Mini-Imagenet 5-way (5-shot) PT+MAP Accuracy 88.82 # 13
Few-Shot Image Classification Mini-ImageNet-CUB 5-way (1-shot) PT+MAP Accuracy 62.49 # 3
Few-Shot Image Classification Mini-ImageNet-CUB 5-way (5-shot) PT+MAP Accuracy 76.51 # 3
Few-Shot Image Classification Tiered ImageNet 5-way (1-shot) PT+MAP Accuracy 85.41 # 4
Few-Shot Image Classification Tiered ImageNet 5-way (5-shot) PT+MAP Accuracy 90.44 # 5

Methods


No methods listed for this paper. Add relevant methods here