Multi-Level Relation Learning with Confidence Evaluation for Few-Shot Learning
Qingjie Zeng, Jie Geng, Kai Huang, Wen Jiang · 2021
Few-shot learning is developed to classify unknown categories through limited training samples. In this paper, a multi-level relation learning model with confidence evaluation (MLR-CE) is proposed in area of few-shot learning. In the proposed framework, multi-level features are extracted that contain semantic information of different depth, and multi-level relation pairs are built by stacking feature maps of support images and query images. To expand the support set, confidence evaluation by a Gaussian mixture model is developed to select samples with high confidences. Experiments on two data demonstrate that the proposed method can yield superior few-shot classification results.