NOTE: This doesn't work at the moment! I'll leave it up because the Releases section has a smaller dataset that I've made available, but don't bother looking at the code.
Train and evaluate a model on whether or not two paintings are by the same painter for the Painter by Numbers challenge on Kaggle.
Uses ResNet50 to learn a distance metric model for whether two paintings are the same.
Preprocessed data used to train and evaluate this model can also be found under the releases tab.
- Better data augmentation/preprocessing?
- Centering/whitening, rotations, etc.