解决少样本问题的策略（可以快速部署）
Damn ·Today topic: For Engineering contribution, you just need to meet the lowest requirement. However, for the graduation project, you have to do huge amount of researches to find one task which worthing to be discussed. Then, give several protential solutions with a bit of corresponding results.
更改链接：
- 使用leaf audio来训练deepship数据来提取模型的特征。生成图片。（任务大船/小船）。
- 对于图片使用小样本学习的策略来识别。孪生网络。
如何 证明mae不适合这个少样本分类网络?
- 跑一下mae的fine tune结果(复现层面)。
- 根据文章 A MUTUAL LEARNING FRAMEWORK FOR FEW-SHOT SOUND EVENT DETECTION，尝试搭建一个少样本分类网络。