Closed wufj closed 3 years ago
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.
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
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
非常谢谢,我去试试!
You are welcome~
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