This is the code you need to train a resnet model and submit to the Human Protein Atlas competition. It uses the newest version (v1) of the fastai library.
To download the data, I used the Official Kaggle API package, which you can install with pip. Once installed, you can run kaggle competitions download -c human-protein-atlas-image-classification to get the data (just make sure to update the path variable in the resnet50_basic.ipynb notebook to point to the data on your machine).
Update 11/17 -- the resnet50_basic notebook doesn't work with fastai version 1.0.25 and above, so I made another notebook to work with the new data_block API. In this version, I also made changes to use the create_cnn function. You can find it at resnet50_basic_datablocks.ipynb.