Train on different dataset

[bhadresh74]

Hi @xingjunm ,

I read your paper and its really intuitive. Thank you for sharing the code. Is it possible to use this code to train a different dataset than you mentioned in your repo. My dataset contains images with noisy multi-label tags and I'd like to see how dimensionality driven learning can improve performance compare to conventional classification models.

Thanks, Bhadresh

[xingjunm]

Hi bhadresh74, thanks for your interest. You can wrap up your new dataset in the datasets.py, load the data the same way as in train_models.py, you can try with a new CNN model (in models.py) or use the resnet base like in the current CIFAR-100 model.