Cross Attention Network for Few-shot Classification

By Rui Hou, Hong Chang, Bingpeng Ma, Shiguang Shan, Xilin Chen. 2019

806 citations recorded by the source corpus.

Few-shot classification aims to recognize unlabeled samples from unseen classes given only few labeled samples. The unseen classes and low-data problem make few-shot classification very challenging. Many existing approaches extracted features from labeled and unlabeled samples independently, as a result, the features are not discriminative enough. In this work, we propose a novel Cross Attention Network to address the challenging problems in few-shot classification. Firstly, Cross Attention Module is introduced to deal with the problem of unseen classes. The module generates cross attention maps for each pair of class feature and query sample feature so as to highlight the target object regions, making the extracted feature more discriminative. Secondly, a transductive inference algorithm is proposed to alleviate the low-data problem, which iteratively utilizes the unlabeled query set to augment the support set, thereby making the class features more representative. Extensive experiments on two benchmarks show our method is a simple, effective and computationally efficient framework and outperforms the state-of-the-arts.

Subfields: Artificial Intelligence, Radiology, Nuclear Medicine and Imaging

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Overcoming Catastrophic Forgetting in Incremental Few-Shot Learning by Finding Flat Minima (2021, methodology): “Current few-shot methods mainly include optimization-based methods [12, 23, 32, 39, 45, 46, 55] and metric-based methods [13, 19, 44, 49, 52, 58, 57, 53].”
Few-Shot and Continual Learning with Attentive Independent Mechanisms (2021, methodology): “As the architecture proposed by [33] is restrictive, [19] proposed a more general approach that uses a cross attention module that models semantic relevance between the support and query set.”
Few-Shot Incremental Learning with Continually Evolved Classifiers (2021, methodology): “Optimization-based methods [8,16,24,28,29,36,37,50] and metric-based methods [9, 12, 35, 42, 46, 47, 51, 52] are two main lines of efforts.”
Meta-Learning in Neural Networks: A Survey (2020, methodology): “For example, they have been used as comparators of support and target set items for metric-based transductive meta-learners [127], as well as feature extractors to prevent catastrophic forgetting in few-shot continual learning [128].”
Efficient Attention: Attention with Linear Complexities (2018, background): “Subsequent works applied it to various fields in computer vision, including image restoration [16], video person re-identification [14], generative adversarial image modeling [28, 2], image de-raining [13], and few-shot learning [9, 11], etc.”
XtarNet: Learning to Extract Task-Adaptive Representation for Incremental Few-Shot Learning (2020, neutral): “46% Cross Attention Nets* (Hou et al., 2019) ResNet12 63.”
An Ensemble of Epoch-Wise Empirical Bayes for Few-Shot Learning (2019, methodology): “It has been applied to many recent works [24,25,45].”
Interventional Few-Shot Learning (2020, methodology): “FSL has a wide spectrum of methods, including fine-tuning [11, 16], optimizing model initialization [18, 40], generating model parameters [51, 34], learning a feature space for a better separation of sample categories [62, 72], feature transfer [54, 41], and transductive learning that additionally u”
A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities (2022, background): “Recently, many works [96, 97] explore useful tricks, including selfsupervised learning [96], attention module [98, 99], and mask generation module [97].”
Research Progress on Few-Shot Learning for Remote Sensing Image Interpretation (2021, background): “natural images, including image classification [17], [18], object detection [19], [20], semantic segmentation [21], [22], etc.”
Learning to Imagine: Integrating Counterfactual Thinking in Neural Discrete Reasoning (2022, background): “Inspired by the success of cross-attention (Kim et al., 2018) in associating different sources, e.g., image-image (Hou et al., 2019) and image-text (Lu et al., 2019), we adopt cross-attention between the factual representation and the assumption representation, followed by self-attention…”
Fine-Grained Few-Shot Classification with Feature Map Reconstruction Networks (2020, methodology): “While prior methods based on feature map reconstruction exist, these methods either rely on constrained iterative procedures [31] or large learned attention modules [5, 9].”
Generalizing Face Forgery Detection with High-frequency Features (2021, background): “[21] derives correlation-based attention for paired class and query features.”
Few-Shot Object Detection and Viewpoint Estimation for Objects in the Wild (2020, methodology): “Rela-tionNet [23] proposes to learn the relation between support data and query data through a neural network, which is similar to CAN [29] and LGM-Net [30].”
DeepEMD: Differentiable Earth Mover's Distance for Few-Shot Learning (2020, methodology): “ods [1], [2], [3], [4], [5], [6], [59], [60], [61], [62], [63], [64],”