A PyTorch-based library for working with 3D and 2D convolutional neural networks, with focus on semantic segmentation of volumetric biomedical image data
The new transforms module contains "non-geometric" augmentations and provides a normalization helper. The general idea of this is very similar to the idea of torchvision.transforms, but based on NumPy and with additional support for joint input-target transforms (see the commit message of 8a89d46 for further details and discussion about API design).
Now it should be very easy to implement new augmentation methods inside and outside of elektronn3.
"Geometric" transforms (source coordinate transformations) will be worked on in the next weeks.
The new
transformsmodule contains "non-geometric" augmentations and provides a normalization helper. The general idea of this is very similar to the idea oftorchvision.transforms, but based on NumPy and with additional support for joint input-target transforms (see the commit message of 8a89d46 for further details and discussion about API design). Now it should be very easy to implement new augmentation methods inside and outside of elektronn3. "Geometric" transforms (source coordinate transformations) will be worked on in the next weeks.This is an important step for #2.