This repo documents how to use fastdup to clean and curate a visual dataset, DVC to version data, and fastai to train a model.
data/ - Stores the train, validation and test images. Tracked by DVC.
models/ - Stores the model trained with fastai. Tracked by DVC.
fastdup_report - Stores the artifacts from running fastdup. Tracked by Git.
DVC versions the data and model produced and syncs them a Google Drive folder.
Initialize dvc repo
dvc init
Add data to watch
dvc add data/
dvc add models/
Add remote Google Drive
dvc remote add -d visuallayergdrive gdrive://1u3nWN53Tp1ZqDEc5VFimu86EjpJ13zrk
Checkout dataset
git checkout <..>
dvc checkout
Push dataset
dvc push