Few-shot learning for image classification task aims to classify images from several novel classes with limited number of samples. Recent studies have shown that the deep local descriptors have better representation ability than image-level features, and achieve great success. However, most of these methods often use all local descriptors or over-screening local descriptors for classification. The former contains some task-irrelevant descriptors, which may misguide the final classification result. The latter is likely to lose some key descriptors. In this paper, we propose a novel Task-Aware Discriminant local descriptors Network (TADNet) to address these issues, which can adaptively select the discriminative query descriptors and eliminate the task-irrelevant query descriptors among the entire task. Specifically, TADNet assigns a value to each query descriptor by comparing its similarity to all support classes to represent its discriminant power for classification. Then the discriminative query descriptors can be preserved via a task-aware attention map. Extensive experiments on both fine-grained and generalized datasets demonstrate that the proposed TADNet outperforms the existing state-of-the-art methods.

This paper proposes an enhanced semi-supervised classification approach termed Nonnegative Sparse Neighborhood Propagation (SparseNP) that is an improvement to the existing neighborhood propagation due to the fact that the outputted soft labels of points cannot be ensured to be sufficiently sparse, discriminative, robust to noise and be probabilistic values. Note that the sparse property and strong discriminating ability of predicted labels is important, since ideally the soft label of each sample should have only one or few positive elements (that is, less unfavorable mixed signs are included) deciding its class assignment. To reduce the negative effects of unfavorable mixed signs on the learning performance, we regularize the l2,1-norm on the soft labels during optimization for enhancing the prediction results. The non-negativity and sum-to-one constraints are also included to ensure the outputted labels are probabilistic values. The proposed framework is solved in an alternative manner for delivering a more reliable solution so that the accuracy can be improved. Simulations show that satisfactory results can be obtained by the proposed SparseNP compared with other related approaches.