More: Writing a custom Dataset

Process of making a new custom dataset is similar to that of PyTorch. Relevant classes are bioimageloader.base.DatasetInterface, bioimageloader.base.Dataset, and bioimageloader.base.MaskDataset. Each class has its own requirements.

Abstract class bioimageloader.base.DatasetInterface defines a basic structure of every dataset classes implemented in bioimageloader. You DO NOT want to make a subclass inherited directly from it.

Instead, you may want to make your subclass based on bioimageloader.base.Dataset or on bioimageloader.base.MaskDataset (more to come). As you might have guessed, class MaskDataset is a base for all datasets that have mask annotation. Class Dataset is a base for those that do not have any annotation available as well as a super class for MaskDataset. In short, if your dataset you would like to implement has mask annotation, then do subclassing from MaskDataset, otherwise do from Dataset.

There is an example template template.py in the git repository. The best practice is to read source codes of existing collections and to see how they are implemented.

You are always welcome to file an issue through Github repository if you need any help.