example deep learning pipeline for OSCER written in Tensorflow
using the pipeline with your own models, datasets, and data augmentation strategies is very simple. One only needs to follow a few conventions:
using your own models: (see supervised/models/cnn.py)
- network functions should only take int, float, or string typed arguments
- network functions must return only a compiled keras model
using your own datasets: (see supervised/datasets/image_classification.py)
- dataset functions should return only an unbatched finite tf.data.Dataset object
using your own data augmentation strategies: (see supervised/data_augmentation/ssda.py)
- data augmentation functions must require only as input a dataset
- data augmentation functions must return only a tf.data.Dataset object representing the augmented dataset
see test.py for an example use case for supervised image classification