The difference between the twodatasets comes from how CIFAR100 is split into meta-train / meta-validation / meta-test sets. Gradients have more noise due to less data variations, compared to higher resolution of miniImageNet images. The state-of-the-art method is shown to outperform ALFA+fo-Proto-MAML in Table D. Inthis section, we study howrobust the proposed meta-learner istochanges indomains, through additional experiments on cross-domain few-shot classification under similar settings to Section 4.3.2 During outer-loop optimization, 15 examples are sampled per each class forD0. All models were trained for 50000 iterations with the meta-batch size of 2 and 4 tasks for 5-shot and 1-shot, respectively.

The conventional few-shot classification aims at learning a model on a large labeled base dataset and rapidly adapting to a target dataset that is from the same distribution as the base dataset. However, in practice, the base and the target datasets of few-shot classification are usually from different domains, which is the problem of cross-domain few-shot classification. We tackle this problem by making a small proportion of unlabeled images in the target domain accessible in the training stage. In this setup, even though the base data are sufficient and labeled, the large domain shift still makes transferring the knowledge from the base dataset difficult. We meticulously design a cross-level knowledge distillation method, which can strengthen the ability of the model to extract more discriminative features in the target dataset by guiding the network's shallow layers to learn higher-level information. Furthermore, in order to alleviate the overfitting in the evaluation stage, we propose a feature denoising operation which can reduce the feature redundancy and mitigate overfitting. Our approach can surpass the previous state-of-the-art method, Dynamic-Distillation, by 5.44% on 1-shot and 1.37% on 5-shot classification tasks on average in the BSCD-FSL benchmark. The implementation code will be available at https://github.com/jarucezh/cldfd.

Abstract: Few-shot classification aims to recognize novel categories with only few labeled images in each class. Existing metric-based few-shot classification algorithms predict categories by comparing the feature embeddings of query images with those from a few labeled images (support examples) using a learned metric function. While promising performance has been demonstrated, these methods often fail to generalize to unseen domains due to large discrepancy of the feature distribution across domains. In this work, we address the problem of few-shot classification under domain shifts for metric-based methods. Our core idea is to use feature-wise transformation layers for augmenting the image features using affine transforms to simulate various feature distributions under different domains in the training stage.