Abstract:Few-shot object detection (FSOD) aims at designing models that can accurately detect targets of novel classes in a scarce data regime. Existing research has improved detection performance with meta-learning-based models. However, existing methods continue to exhibit certain imperfections: (1) Only the interacting global features of query and support images lead to ignoring local critical features in the imprecise localization of objects from new categories. (2) Convolutional neural networks (CNNs) encounter difficulty in learning diverse pose features from exceedingly limited labeled samples of unseen classes. (3) Local context information is not fully utilized in a global attention mechanism, which means the attention modules need to be improved. As a result, the detection performance of novel-class objects is compromised. To overcome these challenges, a few-shot object detection network is proposed with a local feature enhancement module and an intrinsic feature transformation module. In this paper, a local feature enhancement module (LFEM) is designed to raise the importance of intrinsic features of the novel-class samples. In addition, an Intrinsic Feature Transform Module (IFTM) is explored to enhance the feature representation of novel-class samples, which enriches the feature space of novel classes. Finally, a more effective cross-attention module, called Global Cross-Attention Network (GCAN), which fully aggregates local and global context information between query and support images, is proposed in this paper. The crucial features of novel-class objects are extracted effectively by our model before the feature fusion between query images and support images. Our proposed method increases, on average, the detection performance by 0.93 (nAP) in comparison with previous models on the PASCAL VOC FSOD benchmark dataset. Extensive experiments demonstrate the effectiveness of our modules under various experimental settings.

Few-Shot Intent Detection with Label-Enhanced Hierarchical Feature Learning and Graph Neural Networks

Few-shot Learning for Multi-label Intent Detection

Label Hierarchical Structure-Aware Multi-Label Few-Shot Intent Detection Via Prompt Tuning

Dual Class Knowledge Propagation Network for Multi-label Few-shot Intent Detection

A Coarse-to-Fine Prototype Learning Approach for Multi-Label Few-Shot Intent Detection

Few-Shot Object Detection with Local Feature Enhancement and Feature Interrelation

Strengthened Multiple Correlation for Multi-Label Few-Shot Intent Detection

Multi-Label Few-Shot Learning for Aspect Category Detection

Few-shot Intent Detection with Mutual Information and Contrastive Learning

Improving Few-shot Text Classification via Pretrained Language Representations

An Explicit-Joint and Supervised-Contrastive Learning Framework for Few-Shot Intent Classification and Slot Filling

Self-similarity Feature Based Few-Shot Learning Via Hierarchical Relation Network.

Learning to Bridge Metric Spaces: Few-shot Joint Learning of Intent Detection and Slot Filling.

Dynamic Label Name Refinement for Few-Shot Dialogue Intent Classification

Learning to Classify Intents and Slot Labels Given a Handful of Examples

A Contrastive Learning-Based Task Adaptation Model for Few-Shot Intent Recognition

FewJoint: few-shot learning for joint dialogue understanding

HIPPL: Hierarchical Intent-Inferring Pointer Network With Pseudo Labeling for Consistent Persona-Driven Dialogue Generation

Relation Knowledge Distillation Based on Prompt Learning for Generalized Few-Shot Intent Detection

Continual Few-shot Intent Detection.

Few-Shot Audio Classification with Attentional Graph Neural Networks