search

JinyuCai95 / EDESC-pytorch

MIT License
23 stars 6 forks source link

About Data Loading #1

Open V1oletM opened 2 years ago

V1oletM commented 2 years ago

Hi, Thanks for your code. It helps a lot but I'm confused about how to load the data from CIFAR-10 and CIFAR-100. May I have to load their x and y into .npy file? Also, is it available for other datasets? Thanks a lot!

JinyuCai95 commented 2 years ago

Hi, I think you can load arbitrary datasets from various types of files, e.g., .mat, .npz, .npy, .h5 are all acceptable, just make sure the format of dataset is the same as described in utils.py.

V1oletM commented 2 years ago

Is it right if I load the 2048-dimensional features extracted by ResNet-50 mentioned in the paper as x when using image datasets? And the does the parameter n_z have to be equal to the num_classes * d? Thanks!

JinyuCai95 commented 2 years ago

That's right, and n_z has to be equal to num_classes * d, you can refer to InitiallizeD.py.