baby.training¶
Utilities for training the steps of BABY separately or sequentially.
The training module combines all of the utilities and structures needed to train a BABY segmentation framework from scratch. It includes the following trainers * SmoothingModelTrainer: hyper-parameters for smooth data augmentation * FlattenerTrainer: hyper-parameters for distinguishing CNN outputs * HyperParameterTrainer: CNN hyper-parameters * CNNTrainer: CNN using gradient descent to optimize for a given loss * SegmentationTrainer: hyper-parameters for post-processing of CNN Output into cell instances and attributes
Given the appropriate inputs, each of these can be trained separately. This is useful for fine-tuning or re-training parts separately.
For training the entire framework at once, it is recommended to use the BabyTrainer class, which is also aliased as Nursery.
Modules
Optimising the hyper-parameters of the SegmentationFlattener |
|