The classification accuracy of the flower dataset is low

[lucy11111111]

I used the flower classification dataset from this URL for training：https://www.robots.ox.ac.uk/~vgg/data/flowers/102/, but I got low accuracy without modifying any parameter of the original network, may I ask why this? I used some other dataset for training and also got the same low accuracy.

[jameslahm]

Did you load the pre-trained checkpoint?

[lucy11111111]

Isn't this a pre-trained weight in the original code? parser.add_argument('--teacher-path', type=str, default='https://dl.fbaipublicfiles.com/deit/regnety_160-a5fe301d.pth')

[jameslahm]

No, this is the pre-trained weight of the teacher model. I mean the pre-trained weight of the student model on ImageNet.

[lucy11111111]

I see, is it easy to give a download link for the pre-trained weights? Can you tell me exactly how this weight is put into the code?

[jameslahm]

The download links for the pre-trained weights are here: https://github.com/THU-MIG/RepViT#models

[lucy11111111]

May I ask where this weight is loaded in the code? Where do I need to change the code?

[jameslahm]

You could refer to https://pytorch.org/tutorials/beginner/saving_loading_models.html.