openvino_en.md

TNN X86/Openvino Documentation

Introduction to TNN X86/Openvino

This module supports TNN on x86 architecture and includes Openvino framework into TNN, which allows a TNN model running on Openvino Network.

Environment Requirements

Linux

Cmake(>=3.7.2)

Windows

Visual Stuido(>=2017)
CMake(>=3.7.2 or use build-in CMake in Visual Studio Tools)

Compile with scripts

Linux:
$ cd scripts/
$ sh build_linux.sh

Windows:
cd scripts\
.\build_msvc.bat [VS2015/VS2017/VS2019]

Refer to FAQ if failed.

How to run

1. Run with intergrated test file

Move to build_openvino/test/, run TNNTest with model, and set device_type to X86

$ cd build_openvino/test/
$ ./TNNTest -mp PATH_TO_MODEL -dt X86 -ip PATH_TO_INPUT -op PATH_TO_OUTPUT

2. API Documentation

Refer to API Documentation, which needs to set config.device_type as DEVICE_X86 and config.network_type as NETWORK_TYPE_OPENVINO

config.device_type  = TNN_NS::DEVICE_X86
// run with native x86 optimized code, if network type is not set
config.network_type = TNN_NS::NETWORK_TYPE_OPENVINO

Run with demo

Move to example/openivno/ and run build_openvino.sh to compile demos with x86 architecture. Then call demo_x86_linux_imageclassify or demo_x86_linux_facedetector to run demos. For details move to

Run Demo

Refer to demo documentaion

FAQ

Q: CMake not found in Windows?
A: If CMake was installed, add the CMake path to Windows Environment Viraibles. Or use Visual Studio Prompt to run build_x86_msvc.bat, which includes build-in CMake.

Q: Visual Studio not found in Windows?
A: Execute the scripts with Visual Studio Version, Like

.\build_x86_msvc.bat VS2019

Q: Error 0x4001 or 16385 with message "Invalid Model Content"
A: set std::ios::binary when reading Model stream:

std::ifstream model_stream(mode_path, std::ios::binary);