分本文档分类

[稻678]

您好，我看了您的论文，感觉对我们的研究贡献很大，但是有一点疑惑，如果我用贵网络来进行文本文档分类，--data-set是从['CIFAR', 'CIFAR10', 'IMNET', 'INAT', 'INAT19']选择哪个好点呢

[Zhiliang Peng]

Thanks for your kind words! ['CIFAR', 'CIFAR10', 'IMNET', 'INAT', 'INAT19'] are datasets for image classification. If you want to transfer Conformer to your own task, maybe you should define specific dataset.

[稻678]

谢谢您的回复，如果我定义文本文档相关的数据集，是不是得改很多代码呢，还是我指在['CIFAR', 'CIFAR10', 'IMNET', 'INAT', 'INAT19']加一个类就行呢

[Zhiliang Peng]

It depends on your dataset format, if your dataset format is as follows:

/path/to/your dataset/
  train/
    class1/
      img1.jpeg
    class2/
      img2.jpeg
  val/
    class1/
      img3.jpeg
    class/2
      img4.jpeg

Then you just need to add in the build_dataset() of dataset.py:

    elif args.data_set == 'YOUR DATASET NAME':
        root = os.path.join(args.data_path, 'train' if is_train else 'val')
        dataset = datasets.ImageFolder(root, transform=transform)
        nb_classes = XXXX

[稻678]

非常谢谢，我去试试！

[Zhiliang Peng]

You are welcome~