Three-Dimension Attention Mechanism and Self-Supervised Pretext Task for Augmenting Few-Shot Learning

The main challenge of few-shot learning lies in the limited labeled sample data. In addition, since image-level labels are usually not accurate describing features images, it leads to difficulty for model have good generalization ability and robustness. This problem has been well solved yet, existing metric-based methods still room improvement. To address this issue, we propose a method based on three-dimension attention mechanism self-supervised learning. module is used extract more representative by focusing semantically informative through spatial channel attention. Self-supervised mainly adopts proxy task rotation transformation, which increases semantic information without requiring additional manual labeling, uses training combination with supervised loss function improve We conducted extensive experiments four popular datasets achieved state-of-the-art performance both 5-shot 1-shot scenarios. Experiment results show that our work provides novel remarkable approach