vak.transforms.defaults.frame_classification

Default transforms for frame classification models.

These are “item” transforms because they apply transforms to input parameters and then return them in an “item” (dictionary) that is turn returned by the __getitem__ method of a vak.InferDatapipe. Having the transform return a dictionary makes it possible to avoid coupling the InferDatapipe __getitem__ implementation to the transforms needed for specific neural network models, e.g., whether the returned output includes a mask to crop off padding that was added.

Classes

InferItemTransform(window_size[, ...])	Default transform used when running inference on frame classification models, for evaluation or to generate new predictions.
TrainItemTransform([frames_standardizer])	Default transform used when training frame classification models