* [MO] [Kaldi] Added TDNN Component
* TdnnComponent replacer graphical comment updated
* Added SpecAugmentTimeMaskComponent
* some refactor of memoryoffset shape_infer
* moved memoryoffset splitting to the middle stage
* some corrections
- set `need_shape_inferenc`=False in split_memoryoffset
- use cycle instead of pattern in tdnn_replacer
* separated splitting of MemoryOffsets in LSTM and TDNN blocks
* set transpose_weights=True in TdnnComponent
* Corrected Supported_Frameworks_Layers
* corrected comments
* separate naming for tdnn and lstm memoryoffset splits
* corrected BOM file
* corrected generaldropout_ext.py and removed 'has_default' for tdnn_component
* corrections after PR review
* renamed LSTM -> recurrent; added setting element_size for paired nodes of tdnn_memoffset and othe minor changes
* Update split_tdnn_memoryoffset.py
* corrected partial infer with new API in elemental.py and split_tdnn_memoryoffset.py
* Commit.
* Added opset4 version in the class Interpolate.
* Added class ONNXResize11Op to read ONNX Resize with opset version >= 11.
* Added support for Interpolate-4 into transformations TestInterpolateReshapeWA and InterpolateConcat.
* Added support for Interpolate-4 into transformation InterpolateWithConcat.
* Deleted redundant checks from the transformation UpsampleToResample.
* Reverted last changes.
* Changed ONNX Resize extractor to support for Interpolate-4.
* Added conversion of ONNXResize11Op into Interpolate-4.
* Added support for Interpolate-4 into the transformation InterpolateSequenceToInterpolate.
* Small fix for formatting.
* Written tests for MO version of Interpolate-4 with shape_calculation_mode = sizes.
* Written tests for infer function of Interpolate-4.
* Now transformations InterpolateWithConcat, InterpolateConcat, InterpolateReshapeWA skip Interpolate-4.
* Used create_op_with_const_inputs in the transformation InterpolateSequenceToInterpolate.
* The transformation ONNXResize11ToInterpolate4 was rewritten using find_and_replace_pattern.
* Now the dictionary infers (dictionary of infer functions of Interpolate) is a class static attribute.
* Deleted unused variable.
* Restored original logic of find_and_replace_pattern method of the class InterpolateReshapeWA.
* Used create_op_with_const_inputs() in the transformation InterpolateSequenceToInterpolate for opset1 case.
* Replaced resize_name by resize.soft_get('name', resize.id).
* Small fixes.
* Added two tests for Interpolate-4 infer function.
* Fixed the transformation ONNXResize11ToInterpolateV4 for the case when ONNXResize11 operation has 3 inputs.
* Added conversion of ONNXResize11 with tf_crop_and_resize_mode to ROIPooling + ONNXResize11.
* Fixed bugs in the transformation ONNXResize11ToInterpolateV4 and in the infer function of the operation ONNXResize11.
* Small changes.
* Renamed transformation that converts ONNXResize11 into ROIPooling + ONNXResize11 and fixed BOM-file.
* Fixed tests for the transformation InterpolateSequenceToInterpolate.
* Small change.
* Now the transformation InterpolateSequenceToInterpolate preserves output layer name.
* Deleted the transformation ONNXResize11ToTFCropAndResize.
* Fix fusing Multiply node with Convolution in case group != 1
* Add transformation test
* Do not fuse if not possible to reshape const
* Update fuse_linear_ops.py
* Updated ConcatOptimization transformation to work when one dimension of input to Concat is 0D
* Fixed ConcatOptimization transformation to reconnect input edges to Concat
* Completely re-written ConcatOptimization
* Updated Concat0D optimization transformation
* Fixed order of traversing Concat input ports
* Refactored ConcatOptimization transformation to use `delete_input_port` function
* Detele trailing unconnected ports in the ConcatOptimization.py
* Cleaner implementation of ConcatOptimization + unit test
* initial commit
* first reshap-able variant
* right version for reshape
* comment update
* fixes for failed e2e
* set data type to ngraph TensorIterator
* Fix dynamic shapes for cells ops
* clean up
Co-authored-by: yegor.kruglov <ykruglov@nnlvdp-mkaglins.inn.intel.com>
* Implement reshapeable CTCGreedyDecoderPlusSparseToDense transformation and test
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Fix consts (after code-review #1)
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Add CTCGreedyDecoderTransformation with more generic pattern
Also it adds new middle-replacer for transforming sequence length to a mask
along with tests.
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Do fixes after review #2
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Fix after review #3
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Fix after review #4
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Added HSwish operation
* Added HSwish fusing transformation
* Fixed BOM
* Added unit test for HSwish fusing transformation
* Fixed unit tests for transformations using 'build_graph_with_edge_attrs' function to build the graph
* Added fusion transformation for Swish operation
* Added fusing transformation for Softplus operation
* Added fusion transformation for Mish operation
* Added check for the node name in the unit tests
* Fixed Mish fusion pattern
* Updated Mish fusion transformation. Added unit test
* Updated HSwish fusing transformation
* Updated Swish fusion transformation and tests
* Fixed unit tests
* Fixed order of transformation to convert the TF OD API SSD models
* Refactored the sub-graph modification for the TF OD API models related to Squeeze/Reshape after SSD heads
* Extend MO for operation CTCLoss
* Change sequence length format to a mask format
* Add fixes after first-round review
* Add fixes after the second-round review
* Fixing CTCLossPlusCTCGreedyDecoder transformation
