dlc2action.transformer

Augmentations and feature concatenation

Input data in dlc2action is stored in feature dictionaries right up until being passed to a model. That makes it possible to define universal augmentations and SSL transformations that work on a range of datasets naturally. Those dictionaries are generated by a feature extractor when a dlc2action.data.base_store.InputStore instance is initialised. They are then (optionally) processed by SSL transformations (see dlc2action.ssl) and finally merged into tensors by *transformers*.dlc2action.transformer.base_transformer.Transformer` instances can also perform augmentations before the merging.

View Source

 1#
 2# Copyright 2020-2022 by A. Mathis Group and contributors. All rights reserved.
 3#
 4# This project and all its files are licensed under GNU AGPLv3 or later version. A copy is included in dlc2action/LICENSE.AGPL.
 5#
 6"""
 7## Augmentations and feature concatenation
 8
 9Input data in `dlc2action` is stored in feature dictionaries right up until being passed to a model.
10That makes it possible to define universal augmentations and SSL transformations that work on a range of datasets
11naturally. Those dictionaries are generated by a *feature extractor* when a `dlc2action.data.base_store.InputStore`
12instance is initialised. They are
13then (optionally) processed by SSL transformations (see `dlc2action.ssl) and finally merged into tensors by
14*transformers*. `dlc2action.transformer.base_transformer.Transformer` instances
15can also perform augmentations before the merging.
16"""
17
18from dlc2action.transformer.kinematic import KinematicTransformer