Your generously provided code helps me a lot for implementing new thang from existing methods.
Having said that, existing code do not show the reported performances of DM, even when i reproduced it with same hyper-parameters.
In detail, results for CIFAR10 with 50 samples per classes does not go over 55% even when i try bunch of hyper-parameter tunings.
Your generously provided code helps me a lot for implementing new thang from existing methods. Having said that, existing code do not show the reported performances of DM, even when i reproduced it with same hyper-parameters. In detail, results for CIFAR10 with 50 samples per classes does not go over 55% even when i try bunch of hyper-parameter tunings.