Add Arm NN support

NOTICE.md

@@ -40,6 +40,10 @@ This project utilizes the following OSS (Open Source Software):
    - https://developer.qualcomm.com/software/qualcomm-neural-processing-sdk
    - Copyright (c) 2017-2020 Qualcomm Technologies, Inc.
 
+- Arm NN
+   - https://github.com/Arm-software/armnn
+   - Copyright (c) 2017 ARM Limited.
+
 --------------------------------------------------------------------
 Terms of the Apache License, Version 2.0:
 
@@ -270,3 +274,28 @@ Permission is hereby granted, free of charge, to any person obtaining a copy of
 The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
 
 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+
+--------------------------------------------------------------------
+MIT License
+
+Copyright (c) 2017 ARM Limited.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+

README.md

@@ -12,6 +12,7 @@
 - ncnn
 - MNN
 - SNPE (Snapdragon Neural Processing Engine SDK (Qualcomm Neural Processing SDK for AI v1.51.0))
+- Arm NN
 
 ## Supported targets
 - Windows 10 (Visual Studio 2017 x64, Visual Studio 2019 x64)
@@ -31,6 +32,7 @@
 | ncnn                      | OK                       | OK            | OK            | OK               | OK                |
 | MNN                       | OK                       | OK            | OK            | OK               | OK                |
 | SNPE                      | not supported            | not supported | not tested    | OK               | OK                |
+| Arm NN                    | not supported            | OK            | not supported | OK               | not supported     |
 | Note                      | Visual Studio 2017, 2019 | Xubuntu 18.04 | Raspberry Pi  | Jetson Xavier NX | Pixel 4a          |
 
 
@@ -42,7 +44,6 @@ https://github.com/iwatake2222/InferenceHelper_Sample
 - https://github.com/iwatake2222/play_with_tensorrt
 - https://github.com/iwatake2222/play_with_ncnn
 - https://github.com/iwatake2222/play_with_mnn
-- https://github.com/iwatake2222/play_with_snpe
 
 # Usage
 ## Installation
@@ -103,6 +104,8 @@ https://github.com/iwatake2222/InferenceHelper_Sample
     cmake .. -DINFERENCE_HELPER_ENABLE_MNN=on
     # SNPE
     cmake .. -DINFERENCE_HELPER_ENABLE_SNPE=on
+    # Arm NN
+    cmake .. -DINFERENCE_HELPER_ENABLE_ARMNN=on
     ```
 
 - Enable/Disable preprocess using OpenCV:
@@ -127,6 +130,7 @@ typedef enum {
     kNcnn,
     kMnn,
     kSnpe,
+    kArmnn,
 } HelperType;
 ```
 
@@ -175,15 +179,15 @@ InputTensorInfo input_tensor_info("input", TensorInfo::TENSOR_TYPE_FP32);
 input_tensor_info.tensor_dims = { 1, 224, 224, 3 };
 input_tensor_info.data_type = InputTensorInfo::kDataTypeImage;
 input_tensor_info.data = img_src.data;
-input_tensor_info.imageInfo.width = img_src.cols;
-input_tensor_info.imageInfo.height = img_src.rows;
-input_tensor_info.imageInfo.channel = img_src.channels();
-input_tensor_info.imageInfo.cropX = 0;
-input_tensor_info.imageInfo.cropY = 0;
-input_tensor_info.imageInfo.cropWidth = img_src.cols;
-input_tensor_info.imageInfo.cropHeight = img_src.rows;
-input_tensor_info.imageInfo.isBGR = false;
-input_tensor_info.imageInfo.swapColor = false;
+input_tensor_info.image_info.width = img_src.cols;
+input_tensor_info.image_info.height = img_src.rows;
+input_tensor_info.image_info.channel = img_src.channels();
+input_tensor_info.image_info.crop_x = 0;
+input_tensor_info.image_info.crop_y = 0;
+input_tensor_info.image_info.crop_width = img_src.cols;
+input_tensor_info.image_info.crop_height = img_src.rows;
+input_tensor_info.image_info.is_bgr = false;
+input_tensor_info.image_info.swap_color = false;
 input_tensor_info.normalize.mean[0] = 0.485f;   /* https://github.com/onnx/models/tree/master/vision/classification/mobilenet#preprocessing */
 input_tensor_info.normalize.mean[1] = 0.456f;
 input_tensor_info.normalize.mean[2] = 0.406f;
@@ -228,6 +232,7 @@ inference_helper->Process(output_tensor_info_list)
 enum {
     kTensorTypeNone,
     kTensorTypeUint8,
+    kTensorTypeInt8,
     kTensorTypeFp32,
     kTensorTypeInt32,
     kTensorTypeInt64,
@@ -288,7 +293,7 @@ struct {
 void* data;     // [Out] Pointer to the output data_
 struct {
     float   scale;
-    uint8_t zeroPoint;
+    uint8_t zero_point;
 } quant;        // [Out] Parameters for dequantization (convert uint8 to float)
 ```
 

inference_helper/inference_helper.h

@@ -135,7 +135,7 @@ class OutputTensorInfo : public TensorInfo {
 #pragma omp parallel
             for (int32_t i = 0; i < data_num; i++) {
                 const uint8_t* val_uint8 = static_cast<const uint8_t*>(data);
-                float val_float = (val_uint8[i] - quant.zeroPoint) * quant.scale;
+                float val_float = (val_uint8[i] - quant.zero_point) * quant.scale;
                 data_fp32_[i] = val_float;
             }
             return data_fp32_;
@@ -150,7 +150,7 @@ class OutputTensorInfo : public TensorInfo {
     void* data;     // [Out] Pointer to the output data_
     struct {
         float   scale;
-        uint8_t zeroPoint;
+        uint8_t zero_point;
     } quant;        // [Out] Parameters for dequantization (convert uint8 to float)
 
 private:

inference_helper/inference_helper_armnn.cpp

@@ -207,7 +207,7 @@ class ArmnnWrapper {
             list_armnntensor_out_.push_back(std::make_pair(armnn_info.first, armnn::Tensor(armnn_info.second, list_buffer_out_.back())));
 
             tensor_info.data = list_buffer_out_.back();
-            tensor_info.quant.zeroPoint = armnn_tensor_info.GetQuantizationOffset();
+            tensor_info.quant.zero_point = armnn_tensor_info.GetQuantizationOffset();
             tensor_info.quant.scale = armnn_tensor_info.GetQuantizationScale();
         }
 

inference_helper/inference_helper_tensorflow_lite.cpp

@@ -363,7 +363,7 @@ void InferenceHelperTensorflowLite::DisplayModelInfo(const tflite::Interpreter&
         }
         if (tensor->type == kTfLiteUInt8) {
             PRINT("    tensor[%d]->type: quantized\n", i);
-            PRINT("    tensor[%d]->params.outputZeroPoint, scale: %d, %f\n", i, tensor->params.zero_point, tensor->params.scale);
+            PRINT("    tensor[%d]->params.zero_point, scale: %d, %f\n", i, tensor->params.zero_point, tensor->params.scale);
         } else {
             PRINT("    tensor[%d]->type: not quantized\n", i);
         }
@@ -380,7 +380,7 @@ void InferenceHelperTensorflowLite::DisplayModelInfo(const tflite::Interpreter&
         }
         if (tensor->type == kTfLiteUInt8) {
             PRINT("    tensor[%d]->type: quantized\n", i);
-            PRINT("    tensor[%d]->params.outputZeroPoint, scale: %d, %f\n", i, tensor->params.zero_point, tensor->params.scale);
+            PRINT("    tensor[%d]->params.zero_point, scale: %d, %f\n", i, tensor->params.zero_point, tensor->params.scale);
         } else {
             PRINT("    tensor[%d]->type: not quantized\n", i);
         }
@@ -468,7 +468,7 @@ int32_t InferenceHelperTensorflowLite::GetOutputTensorInfo(OutputTensorInfo& ten
                 tensor_info.tensor_type = TensorInfo::kTensorTypeUint8;
                 tensor_info.data = interpreter_->typed_tensor<uint8_t>(i);
                 tensor_info.quant.scale = tensor->params.scale;
-                tensor_info.quant.zeroPoint = tensor->params.zero_point;
+                tensor_info.quant.zero_point = tensor->params.zero_point;
                 break;
             case kTfLiteFloat32:
                 tensor_info.tensor_type = TensorInfo::kTensorTypeFp32;

inference_helper/inference_helper_tensorrt.cpp

@@ -430,7 +430,7 @@ int32_t InferenceHelperTensorRt::AllocateBuffers(std::vector<InputTensorInfo>& i
                     }
                     if (data_type == nvinfer1::DataType::kINT8) {
                         output_tensor_info.quant.scale = 1.0;			// todo
-                        output_tensor_info.quant.zeroPoint = 0.0;
+                        output_tensor_info.quant.zero_point = 0.0;
                     }
                     output_tensor_info.data = buffer_cpu;
                 }