diff options
author | Kshitij Sisodia <kshitij.sisodia@arm.com> | 2022-05-06 09:13:03 +0100 |
---|---|---|
committer | Kshitij Sisodia <kshitij.sisodia@arm.com> | 2022-05-06 17:11:41 +0100 |
commit | aa4bcb14d0cbee910331545dd2fc086b58c37170 (patch) | |
tree | e67a43a43f61c6f8b6aad19018b0827baf7e31a6 /source/application/api/use_case/vww | |
parent | fcca863bafd5f33522bc14c23dde4540e264ec94 (diff) | |
download | ml-embedded-evaluation-kit-aa4bcb14d0cbee910331545dd2fc086b58c37170.tar.gz |
MLECO-3183: Refactoring application sources
Platform agnostic application sources are moved into application
api module with their own independent CMake projects.
Changes for MLECO-3080 also included - they create CMake projects
individial API's (again, platform agnostic) that dependent on the
common logic. The API for KWS_API "joint" API has been removed and
now the use case relies on individual KWS, and ASR API libraries.
Change-Id: I1f7748dc767abb3904634a04e0991b74ac7b756d
Signed-off-by: Kshitij Sisodia <kshitij.sisodia@arm.com>
Diffstat (limited to 'source/application/api/use_case/vww')
5 files changed, 304 insertions, 0 deletions
diff --git a/source/application/api/use_case/vww/CMakeLists.txt b/source/application/api/use_case/vww/CMakeLists.txt new file mode 100644 index 0000000..b933d32 --- /dev/null +++ b/source/application/api/use_case/vww/CMakeLists.txt @@ -0,0 +1,39 @@ +#---------------------------------------------------------------------------- +# Copyright (c) 2022 Arm Limited. All rights reserved. +# SPDX-License-Identifier: Apache-2.0 +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +#---------------------------------------------------------------------------- +######################################################### +# VISUAL WAKE WORD API library # +######################################################### +cmake_minimum_required(VERSION 3.15.6) + +set(VWW_API_TARGET vww_api) +project(${VWW_API_TARGET} + DESCRIPTION "Visual wake word use case API library" + LANGUAGES C CXX) + +# Create static library +add_library(${VWW_API_TARGET} STATIC + src/VisualWakeWordProcessing.cc + src/VisualWakeWordModel.cc) + +target_include_directories(${VWW_API_TARGET} PUBLIC include) + +target_link_libraries(${VWW_API_TARGET} PUBLIC common_api) + +message(STATUS "*******************************************************") +message(STATUS "Library : " ${VWW_API_TARGET}) +message(STATUS "CMAKE_SYSTEM_PROCESSOR : " ${CMAKE_SYSTEM_PROCESSOR}) +message(STATUS "*******************************************************") diff --git a/source/application/api/use_case/vww/include/VisualWakeWordModel.hpp b/source/application/api/use_case/vww/include/VisualWakeWordModel.hpp new file mode 100644 index 0000000..a34b904 --- /dev/null +++ b/source/application/api/use_case/vww/include/VisualWakeWordModel.hpp @@ -0,0 +1,50 @@ +/* + * Copyright (c) 2021 - 2022 Arm Limited. All rights reserved. + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef VISUAL_WAKE_WORD_MODEL_HPP +#define VISUAL_WAKE_WORD_MODEL_HPP + +#include "Model.hpp" + +namespace arm { +namespace app { + + class VisualWakeWordModel : public Model { + + public: + /* Indices for the expected model - based on input tensor shape */ + static constexpr uint32_t ms_inputRowsIdx = 1; + static constexpr uint32_t ms_inputColsIdx = 2; + static constexpr uint32_t ms_inputChannelsIdx = 3; + + protected: + /** @brief Gets the reference to op resolver interface class. */ + const tflite::MicroOpResolver& GetOpResolver() override; + + /** @brief Adds operations to the op resolver instance. */ + bool EnlistOperations() override; + private: + /* Maximum number of individual operations that can be enlisted. */ + static constexpr int ms_maxOpCnt = 7; + + /* A mutable op resolver instance. */ + tflite::MicroMutableOpResolver<ms_maxOpCnt> m_opResolver; + }; + +} /* namespace app */ +} /* namespace arm */ + +#endif /* VISUAL_WAKE_WORD_MODEL_HPP */ diff --git a/source/application/api/use_case/vww/include/VisualWakeWordProcessing.hpp b/source/application/api/use_case/vww/include/VisualWakeWordProcessing.hpp new file mode 100644 index 0000000..f9f9d72 --- /dev/null +++ b/source/application/api/use_case/vww/include/VisualWakeWordProcessing.hpp @@ -0,0 +1,93 @@ +/* + * Copyright (c) 2022 Arm Limited. All rights reserved. + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef VWW_PROCESSING_HPP +#define VWW_PROCESSING_HPP + +#include "BaseProcessing.hpp" +#include "Model.hpp" +#include "Classifier.hpp" + +namespace arm { +namespace app { + + /** + * @brief Pre-processing class for Visual Wake Word use case. + * Implements methods declared by BasePreProcess and anything else needed + * to populate input tensors ready for inference. + */ + class VisualWakeWordPreProcess : public BasePreProcess { + + public: + /** + * @brief Constructor + * @param[in] inputTensor Pointer to the TFLite Micro input Tensor. + * @param[in] rgb2Gray Convert image from 3 channel RGB to 1 channel grayscale. + **/ + explicit VisualWakeWordPreProcess(TfLiteTensor* inputTensor, bool rgb2Gray=true); + + /** + * @brief Should perform pre-processing of 'raw' input image data and load it into + * TFLite Micro input tensors ready for inference + * @param[in] input Pointer to the data that pre-processing will work on. + * @param[in] inputSize Size of the input data. + * @return true if successful, false otherwise. + **/ + bool DoPreProcess(const void* input, size_t inputSize) override; + + private: + TfLiteTensor* m_inputTensor; + bool m_rgb2Gray; + }; + + /** + * @brief Post-processing class for Visual Wake Word use case. + * Implements methods declared by BasePostProcess and anything else needed + * to populate result vector. + */ + class VisualWakeWordPostProcess : public BasePostProcess { + + private: + TfLiteTensor* m_outputTensor; + Classifier& m_vwwClassifier; + const std::vector<std::string>& m_labels; + std::vector<ClassificationResult>& m_results; + + public: + /** + * @brief Constructor + * @param[in] outputTensor Pointer to the TFLite Micro output Tensor. + * @param[in] classifier Classifier object used to get top N results from classification. + * @param[in] model Pointer to the VWW classification Model object. + * @param[in] labels Vector of string labels to identify each output of the model. + * @param[out] results Vector of classification results to store decoded outputs. + **/ + VisualWakeWordPostProcess(TfLiteTensor* outputTensor, Classifier& classifier, + const std::vector<std::string>& labels, + std::vector<ClassificationResult>& results); + + /** + * @brief Should perform post-processing of the result of inference then + * populate classification result data for any later use. + * @return true if successful, false otherwise. + **/ + bool DoPostProcess() override; + }; + +} /* namespace app */ +} /* namespace arm */ + +#endif /* VWW_PROCESSING_HPP */
\ No newline at end of file diff --git a/source/application/api/use_case/vww/src/VisualWakeWordModel.cc b/source/application/api/use_case/vww/src/VisualWakeWordModel.cc new file mode 100644 index 0000000..2d8a125 --- /dev/null +++ b/source/application/api/use_case/vww/src/VisualWakeWordModel.cc @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2021 Arm Limited. All rights reserved. + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include "VisualWakeWordModel.hpp" +#include "log_macros.h" + +const tflite::MicroOpResolver& arm::app::VisualWakeWordModel::GetOpResolver() +{ + return this->m_opResolver; +} + +bool arm::app::VisualWakeWordModel::EnlistOperations() +{ + this->m_opResolver.AddDepthwiseConv2D(); + this->m_opResolver.AddConv2D(); + this->m_opResolver.AddAveragePool2D(); + this->m_opResolver.AddReshape(); + this->m_opResolver.AddPad(); + this->m_opResolver.AddAdd(); + + if (kTfLiteOk == this->m_opResolver.AddEthosU()) { + info("Added %s support to op resolver\n", + tflite::GetString_ETHOSU()); + } else { + printf_err("Failed to add Arm NPU support to op resolver."); + return false; + } + return true; +} diff --git a/source/application/api/use_case/vww/src/VisualWakeWordProcessing.cc b/source/application/api/use_case/vww/src/VisualWakeWordProcessing.cc new file mode 100644 index 0000000..4ae8a54 --- /dev/null +++ b/source/application/api/use_case/vww/src/VisualWakeWordProcessing.cc @@ -0,0 +1,80 @@ +/* + * Copyright (c) 2022 Arm Limited. All rights reserved. + * SPDX-License-Identifier: Apache-2.0 + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include "VisualWakeWordProcessing.hpp" + +#include "ImageUtils.hpp" +#include "VisualWakeWordModel.hpp" +#include "log_macros.h" + +namespace arm { +namespace app { + + VisualWakeWordPreProcess::VisualWakeWordPreProcess(TfLiteTensor* inputTensor, bool rgb2Gray) + :m_inputTensor{inputTensor}, + m_rgb2Gray{rgb2Gray} + {} + + bool VisualWakeWordPreProcess::DoPreProcess(const void* data, size_t inputSize) + { + if (data == nullptr) { + printf_err("Data pointer is null"); + } + + auto input = static_cast<const uint8_t*>(data); + + uint8_t* unsignedDstPtr = this->m_inputTensor->data.uint8; + + if (this->m_rgb2Gray) { + image::RgbToGrayscale(input, unsignedDstPtr, inputSize); + } else { + std::memcpy(unsignedDstPtr, input, inputSize); + } + + /* VWW model pre-processing is image conversion from uint8 to [0,1] float values, + * then quantize them with input quantization info. */ + QuantParams inQuantParams = GetTensorQuantParams(this->m_inputTensor); + + int8_t* signedDstPtr = this->m_inputTensor->data.int8; + for (size_t i = 0; i < this->m_inputTensor->bytes; i++) { + auto i_data_int8 = static_cast<int8_t>( + ((static_cast<float>(unsignedDstPtr[i]) / 255.0f) / inQuantParams.scale) + inQuantParams.offset + ); + signedDstPtr[i] = std::min<int8_t>(INT8_MAX, std::max<int8_t>(i_data_int8, INT8_MIN)); + } + + debug("Input tensor populated \n"); + + return true; + } + + VisualWakeWordPostProcess::VisualWakeWordPostProcess(TfLiteTensor* outputTensor, Classifier& classifier, + const std::vector<std::string>& labels, std::vector<ClassificationResult>& results) + :m_outputTensor{outputTensor}, + m_vwwClassifier{classifier}, + m_labels{labels}, + m_results{results} + {} + + bool VisualWakeWordPostProcess::DoPostProcess() + { + return this->m_vwwClassifier.GetClassificationResults( + this->m_outputTensor, this->m_results, + this->m_labels, 1, true); + } + +} /* namespace app */ +} /* namespace arm */
\ No newline at end of file |