openvino v2024.1.0
- Created
- Updated
Intel® Distribution of OpenVINO™ toolkit is an open-source toolkit for optimizing
and deploying AI inference. It can be used to develop applications and solutions based on deep learning tasks, such as: emulation of human vision, automatic speech recognition, natural language processing, recommendation systems, etc. It provides high-performance and rich deployment options, from edge to cloud
# Usage
The package OpenVINO can be used in cmake scripts as:
find_package(OpenVINO REQUIRED)
target_link_libraries(main PRIVATE openvino::runtime) Note: static build of Intel GPU plugin disables usage of oneDNN for GPU, which is important
for decent performance on discrete GPU devices. If you want to levarage performance of dGPU,
please, consider either option:
- Building with 'VCPKG_LIBRARY_LINKAGE=dynamic'
- Disable Intel CPU plugin, because it uses different oneDNN flavor
# License
Apache-2.0
# Supported Platforms
!uwp & !x86
- ✅ linux
- ✅ osx
- ✅ windows-arm64
- ✅ windows-x64
- ❌
uwp - ❌
windows-x86
# Features
Default features: cpu , gpu , auto , auto-batch , hetero , ir , onnx , paddle , pytorch , tensorflow , tensorflow-lite
# auto default
Enables Auto plugin for inference
Dependencies:
No dependencies.
Host Dependencies:
No dependencies.
# auto-batch default
Enables Auto Batch plugin for inference, useful for throughput mode
Dependencies:
No dependencies.
Host Dependencies:
No dependencies.
# cpu
Enables CPU plugin for inference
Supports: !(windows & arm)
- ✅ linux
- ✅ osx
- ✅ windows-x64
- ❌
windows-arm64
Dependencies:
No dependencies.
Host Dependencies:
No dependencies.
# gpu
Enables GPU plugin for inference
Supports: x64 & !(osx | uwp)
- ✅ linux
- ✅ windows-x64
- ❌
osx - ❌
windows-arm64
Host Dependencies:
2 transitive dependencies:
- vcpkg-cmake (by opencl )
- vcpkg-cmake-config (by opencl )
# hetero default
Enables Hetero plugin for inference
Dependencies:
No dependencies.
Host Dependencies:
No dependencies.
# ir default
Enables IR frontend for reading models in OpenVINO IR format
Dependencies:
No dependencies.
Host Dependencies:
No dependencies.
# npu
NPU Support
Supports: x64 & !(osx | uwp) & !static
- ✅ linux-dynamic
- ✅ windows-x64-dynamic
- ❌
linux-static - ❌
osx - ❌
windows-arm64 - ❌
windows-x64-static
Dependencies:
No dependencies.
Host Dependencies:
No dependencies.
# onnx default
Enables ONNX frontend for reading models in ONNX format
Host Dependencies:
- protobuf
>= v3.21.2
and 19 transitive dependencies:
- expat (by python3 )
- gettext-libintl (by gettext )
libiconv (by gettext-libintl )
Platforms:
!linux- pkgconf (by vcpkg-pkgconfig-get-modules )
- python3 (by onnx )
- vcpkg-cmake (by protobuf )
- vcpkg-cmake-config (by protobuf )
- vcpkg-cmake-get-vars (by libffi )
vcpkg-msbuild (by python3 )
Platforms:
windows- vcpkg-pkgconfig-get-modules (by vcpkg-msbuild )
- vcpkg-tool-meson (by pkgconf )
- zlib (by python3 )
# paddle default
Enables PaddlePaddle frontend for reading models in PaddlePaddle format
Host Dependencies:
- protobuf
>= v3.21.2
and 2 transitive dependencies:
- vcpkg-cmake (by protobuf )
- vcpkg-cmake-config (by protobuf )
# pytorch default
Enables PyTorch frontend to convert models in PyTorch format
Dependencies:
No dependencies.
Host Dependencies:
No dependencies.
# tensorflow default
Enables TensorFlow frontend for reading models in TensorFlow format
Host Dependencies:
- protobuf
>= v3.21.2
and 2 transitive dependencies:
- vcpkg-cmake (by protobuf )
- vcpkg-cmake-config (by protobuf )
# tensorflow-lite default
Enables TensorFlow Lite frontend for reading models in TensorFlow Lite format
Host Dependencies:
- flatbuffers
>= v2.0.6
and 2 transitive dependencies:
- vcpkg-cmake (by flatbuffers )
- vcpkg-cmake-config (by flatbuffers )
# Dependencies
# Host Dependencies
and one transitive dependency:
- vcpkg-tool-meson (by pkgconf )
# Dependents
# Host Dependents
No dependents.
# Contributors
Ilya Lavrenov
Lily Wang
Cheney Wang
JonLiu1993
Billy O'Neal
autoantwort
Tatyana Raguzova
Park DongHa
Egor Tyuvaev