How to train a new dataset?

[mohammad-saber]

Thank you very much for sharing your code.

I want to train models with a dataset other than miniImageNet and Omniglot. Could you guide me how to arrange dataset (main folder , sub-folder), and how to train models on a new dataset?

Thanks a lot.

[jakesnell]

Hi, one simple way may be to check out the Torchmeta package: https://github.com/tristandeleu/pytorch-meta. They support several datasets and have a prototypical networks implementation.

[zhang1hongliang]

@jakesnell , I am a student and a new hand of few-shot learning. thanks very much for sharing your code of prototypical networks.

When using your code I can only use the Omniglot dataset, but I want to use the miniImageNet dataset to test the improved prototypical networks, can you provide the miniImageNet dataset loading file or tell me how to embedd the Torchmeta package into your code framework.

Looking forward to your reply. Thank you very much.

[DaDaDa-FH]

@jakesnell，我是一名学生，也是少镜头学习的新手。非常感谢您分享原型网络的代码。

使用您的代码时，我只能使用 Omniglot 数据集，但我想使用 miniImageNet 数据集来测试改进的原型网络，您能否提供 miniImageNet 数据集加载文件或告诉我如何将 Torchmeta 包嵌入到您的代码框架中。

期待你的答复。非常感谢。

同学，你解决了吗？我也不会加载miniimagenet数据集

[zhang1hongliang]

@DaDaDa-FH 用这个全监督小样本学习库来实现吧，https://github.com/RL-VIG/LibFewShot
这个你只需要实现你的方法就可以了，数据集加载已经做好了。

[DaDaDa-FH]

@DaDaDa-FH用这个完整的小监督样本学习库来实现吧，https://github.com/RL-VIG/LibFewShot 这个你只需要实现你的方法就可以了，数据集加载已经完成了。

好滴，我试试谢谢了。搞了好久这个作者的miniimagenet就是搞不了一样的精度、