Improved Few-Shot Visual Classification

Few-shot learning is a fundamental task in computer vision that carries the promise of alleviating the need for exhaustively labeled data. Most few-shot learning approaches to date have focused on progressively more complex neural feature extractors and classifier adaptation strategies, as well as the refinement of the task definition itself. In this paper, we explore the hypothesis that a simple class-covariance-based distance metric, namely the Mahalanobis distance, adopted into a state of the art few-shot learning approach (CNAPS) can, in and of itself, lead to a significant performance improvement. We also discover that it is possible to learn adaptive feature extractors that allow useful estimation of the high dimensional feature covariances required by this metric from surprisingly few samples. The result of our work is a new "Simple CNAPS" architecture which has up to 9.2% fewer trainable parameters than CNAPS and performs up to 6.1% better than state of the art on the standard few-shot image classification benchmark dataset.

PDF Abstract CVPR 2020 PDF CVPR 2020 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Few-Shot Image Classification Meta-Dataset Simple CNAPS Accuracy 69.86 # 11
Few-Shot Image Classification Meta-Dataset Rank Simple CNAPS Mean Rank 3.45 # 3
Few-Shot Image Classification Mini-Imagenet 10-way (1-shot) Simple CNAPS Accuracy 37.1 # 6
Few-Shot Image Classification Mini-Imagenet 10-way (1-shot) Simple CNAPS + FETI Accuracy 63.5 # 2
Few-Shot Image Classification Mini-Imagenet 10-way (5-shot) Simple CNAPS + FETI Accuracy 83.1 # 2
Few-Shot Image Classification Mini-Imagenet 10-way (5-shot) Simple CNAPS Accuracy 56.7 # 5
Few-Shot Image Classification Mini-Imagenet 5-way (1-shot) Simple CNAPS Accuracy 53.2 # 88
Few-Shot Image Classification Mini-Imagenet 5-way (1-shot) Simple CNAPS + FETI Accuracy 77.4 # 18
Few-Shot Image Classification Mini-Imagenet 5-way (5-shot) Simple CNAPS Accuracy 70.8 # 79
Few-Shot Image Classification Mini-Imagenet 5-way (5-shot) Simple CNAPS + FETI Accuracy 90.3 # 10
Few-Shot Image Classification Tiered ImageNet 10-way (1-shot) Simple CNAPS Accuracy 48.1 # 4
Few-Shot Image Classification Tiered ImageNet 10-way (1-shot) Simple CNAPS + FETI Accuracy 57.1 # 2
Few-Shot Image Classification Tiered ImageNet 10-way (5-shot) Simple CNAPS Accuracy 70.2 # 4
Few-Shot Image Classification Tiered ImageNet 10-way (5-shot) Simple CNAPS + FETI Accuracy 78.5 # 2
Few-Shot Image Classification Tiered ImageNet 5-way (1-shot) Simple CNAPS + FETI Accuracy 71.4 # 29
Few-Shot Image Classification Tiered ImageNet 5-way (1-shot) Simple CNAPS Accuracy 63.0 # 44
Few-Shot Image Classification Tiered ImageNet 5-way (5-shot) Simple CNAPS + FETI Accuracy 86.0 # 29
Few-Shot Image Classification Tiered ImageNet 5-way (5-shot) Simple CNAPS Accuracy 80.0 # 44

Methods


No methods listed for this paper. Add relevant methods here