The problem of the accuracy

[xuanbinh-nguyen96]

Hi authors,

Thank you for your great work, I have found that the repo very useful and successfully benchmarking in the LIPS (FEMNIST, Sentiment, bla bla bla điền đủ vào) dataset. However, we have got a tiny problem, we are currently struggling when benchmarking the iNaturalist dataset. T(by a large gap). We wonder the models trained in dataset generated by your code at: (naturalist/split_data.py) can not achieve the results as reported in your paper at if the codes for generating iNat data is working well or not. We are looking forward to your help.

Thanks much!