allencell / aics-ml-segmentation Goto Github PK

add a new model similar to unet_xy_zoom, but with zero padding in all conv layer so that the input and ourput have the same size

MONAI (https://github.com/Project-MONAI/MONAI) implements a large number of state-of-the-art models and loss functions. Taking advantage of the MONAI API, we can easily support more models and loss functions

re-implement the inference function for faster sliding window operation (with module from MONAI)

after all key functionalities are implemented, add documentation for how to use these new features

use TorchIO (https://github.com/fepegar/torchio) for 3D transformation (e.g., for data augmentation)

adopt pytorch lightning as the foundation for (1) easy multiple-GPU training/inference, (2) standard boilerplate, (3) more funality in pytorch lighnting making training more efficent (e.g., stochastic weight averaging, automatic learning rate finder, adaptive batch size, etc.)