Repo containing implementations of pretrained networks on various datasets found on Kaggle
To follow and to replicate, you'll need to have a conda/virtualenv setup, with PyTorch >= 0.4, fastai >= 1.0 and all other dependencies. Most of the notebooks utilize the fast.ai library for fast training, and have then been converted to PyTorch models and served up as Heroku web apps
To start, setup the Kaggle CLI according to the instructions described here. This should let you download the datasets with ease.