My Training and Inference Kernel for the Nishika Competition " Challenge machine classification to face classification of people drawn in Japanese paintings!"
My 11th Place solution was using an average Ensemble of Efficient B7 (Noisy weights) and the different variants of Big Transfer based models. I used Mixup and some standard Augmentations like cutout. On top of that I used a Cyclic Learning Rate for better results.
Hope my notebooks can help in coming up with some awesome ideas of your own.