Mltu

Changed
- Changed `mltu.transformers.SpectrogramPadding` object, to pad spectrogram end with zeros instead of start

Added
- Created `Tutorials/09_translation_transformer` tutorial, that shows how to train translation transformer model
- Created `mltu.tensorflow.tokenizers` module, that contains `CustomTokenizer` for text data
- Created `mltu.tensorflow.transformer.attention` module, that contains `BaseAttention`, `CrossAttention`, `GlobalSelfAttention` and `CausalSelfAttention` layers
- Created `mltu.tensorflow.transformer.layers` module, that contains `positional_encoding` function, `PositionalEmbedding`, `FeedForward`, `EncoderLayer`, `DecoderLayer`, `Encoder`, `Decoder` layers and `Transformer` model
- Created `mltu.tensorflow.transformer.callbacks` module, that contains `EncDecSplitCallback` callback, to split Transformer model into separate encoder and decoder models
- Created `mltu.tensorflow.transformer.utils` module, that contains `MaskedLoss` loss and `MaskedAccuracy` metric, used for training Transformer models

Changed
- Fixed bug in `mltu.dataProvider.DataProvider` to work with `batch_postprocessors`.

Changed
- Included `augment_annotation` bool option to all `mltu.augmentors` to be able to choose whether to augment annotation or not
- Changed `mltu.augmentors.RandomRotate` to have `staticmethod` of `rotate_image` to be able to use it without creating object

Added
- Added `batch_postprocessor` option to `mltu.dataProvider.DataProvider` to be able to postprocess batch after augmentation

- Removed because of bad build

Changed
- Moved `onnx` and `tf2onnx` import inside `mltu.tensorflow.callbacks.Model2onnx` to avoid import errors when not using this callback
- Removed `onnx` and `tf2onnx` install requirements from global requirements

Added
- Added `RandomMirror` and `RandomFlip` augmentors into `mltu.augmentors`
- Added `u2net` segmentation model into `mltu.tensorflow.models`

Changed
- Downgrade `tf2onnx` and `onnx` versions, they don't work with newest TensorFlow version