Fashion-MNIST is here!

This is a quick post. I just saw Yann LeCunn’s status update about Fashion-MNIST and I have to say it’s pretty interesting!

It is a data-set designed to be an in-place replacement for MNIST, consisting of 10 classes of 28X28 greyscale images except that instead of digits they’re fashion apparels. The creators have argued that it is time that ML researchers moved away from MNIST because it is way too simple and that Fashion-MNIST should now be used as a benchmark for all Machine Learning Algorithms.

I haven’t yet looked at it in detail. Probably the subject matter of the next blog post. Pretty excited to get my hands on it! It sure does look more comprehensible and much more complex than MNIST. You should check it out if you haven’t!

Github Repo