• Stars
    star
    4,625
  • Rank 9,175 (Top 0.2 %)
  • Language
    C++
  • License
    Apache License 2.0
  • Created almost 7 years ago
  • Updated 2 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Tengine is a lite, high performance, modular inference engine for embedded device

简体中文 | English

Tengine

GitHub license GitHub Workflow Status Test Status codecov Language grade: C/C++

简介

TengineOPEN AI LAB 主导开发,该项目实现了深度学习神经网络模型在嵌入式设备上的快速高效部署需求。为实现在众多 AIoT 应用中的跨平台部署,本项目使用 C 语言进行核心模块开发,针对嵌入式设备资源有限的特点进行了深度框架裁剪。同时采用了完全分离的前后端设计,有利于 CPU、GPU、NPU 等异构计算单元的快速移植和部署,降低评估、迁移成本。

Tengine 核心代码由 4 个模块组成:

  • device:NN Operators 后端模块,已提供 CPU、GPU、NPU 参考代码;
  • scheduler:框架核心部件,包括 NNIR、计算图、硬件资源、模型解析器的调度和执行模块;
  • operator:NN Operators 前端模块,实现 NN Operators 注册、初始化;
  • serializer:模型解析器,实现 tmfile 格式的网络模型参数解析。

架构简析

Tengine 架构

快速上手

编译

  • 快速编译 基于 cmake 实现简单的跨平台编译。

示例

  • examples 提供基础的分类、检测算法用例,根据 issue 需求持续更新。
  • 源安装 提供ubuntu系统的apt-get命令行安装和试用,目前支持x86/A311D硬件。

模型仓库

转换工具

  • 预编译版本 :提供 Ubuntu 18.04 系统上预编译好的模型转换工具;
  • 在线转换版本 :基于 WebAssembly 实现(浏览器本地转换,模型不会上传;
  • 源码编译 :建议在服务器或者PC上编译,指令如下:
    mkdir build && cd build
    cmake -DTENGINE_BUILD_CONVERT_TOOL=ON ..
    make -j`nproc`
    

量化工具

  • 源码编译:已开源量化工具源码,已支持 uint8/int8。

速度评估

  • Benchmark 基础网络速度评估工具,欢迎大家更新。

NPU Plugin

  • TIM-VX VeriSilicon NPU 使用指南。

AutoKernel Plugin

  • AutoKernel 是一个简单易用,低门槛的自动算子优化工具,AutoKernel Plugin实现了自动优化算子一键部署到 Tengine 中。

Container

Roadmap

致谢

Tengine Lite 参考和借鉴了下列项目:

License

澄清说明

  • [在线上报功能] 在线上报功能主要目的是了解Tengine的使用信息,信息用于优化和迭代Tengine,不会影响任何正常功能。该功能默认开启,如需关闭,可修改如下配置关闭:(主目录 CMakeLists.txt ) OPTION (TENGINE_ONLINE_REPORT "online report" OFF)

FAQ

技术讨论

More Repositories

1

TengineKit

TengineKit - Free, Fast, Easy, Real-Time Face Detection & Face Landmarks & Face Attributes & Hand Detection & Hand Landmarks & Body Detection & Body Landmarks & Iris Landmarks & Yolov5 SDK On Mobile.
C++
2,386
star
2

AutoKernel

AutoKernel 是一个简单易用,低门槛的自动算子优化工具,提高深度学习算法部署效率。
C++
776
star
3

FaceDetection

C++ project to implement MTCNN, a perfect face detect algorithm, on different DL frameworks. The most popular frameworks: caffe/mxnet/tensorflow, are all suppported now
C++
270
star
4

Caffe-HRT

Heterogeneous Run Time version of Caffe. Added heterogeneous capabilities to the Caffe, uses heterogeneous computing infrastructure framework to speed up Deep Learning on Arm-based heterogeneous embedded platform. It also retains all the features of the original Caffe architecture which users deploy their applications seamlessly.
C++
269
star
5

YSQfastfd

A fast binary library for face detection and face landmark detection in images. No float point operations, especially suit for low cost Arm CPUs, The highest accuracy on FDDB among non deep learning methods
C++
197
star
6

FaceRecognition

This is an implematation project of face detection and recognition. The face detection using MTCNN algorithm, and recognition using LightenenCNN algorithm.
C++
148
star
7

Tengine-Convert-Tools

Tengine Convert Tool supports converting multi framworks' models into tmfile that suitable for Tengine-Lite AI framework.
C++
94
star
8

TengineFactory

Algorithm acceleration landing framework, let you complete the development of algorithm at low cost.eg: Facedetect, FaceLandmark..
C++
90
star
9

TengineGst

TengineGst is a streaming media analytics framework, based on GStreamer multimedia framework, for creating varied complex media analytics pipelines. It ensures pipeline interoperability and provides optimized media, and inference operations using Tengine Toolkit Inference Engine backend, across varied architecture - CPU, iGPU and VPU.
C++
75
star
10

MXNet-HRT

Heterogeneous Run Time version of MXNet. Added heterogeneous capabilities to the MXNet, uses heterogeneous computing infrastructure framework to speed up Deep Learning on Arm-based heterogeneous embedded platform. It also retains all the features of the original MXNet architecture which users deploy their applications seamlessly.
C++
72
star
11

cortex-m-kws

Cortex M KWS example with Tengine Lite.
C
71
star
12

Tengine-app

This is a repository for applications powered by Tengine. Please feel free to try them on your device
C++
56
star
13

ObjectTracker

This object Tracker algorithm is a TLD(Long-term tracker) tracker base on KCF or DSST.
C++
45
star
14

SpeechRecognition

A local auto speech recognition project based on Kaldi and ALSA.
C++
36
star
15

TensorFlow-HRT

Heterogeneous Run Time version of TensorFlow. Added heterogeneous capabilities to the TensorFlow, uses heterogeneous computing infrastructure framework to speed up Deep Learning on Arm-based heterogeneous embedded platform. It also retains all the features of the original TensorFlow architecture which users deploy their applications seamlessly.
C++
36
star
16

CVGesture

It implements detection and recognition to different hand gestures, based on OpenCV 3.3.0 (Open Source Computer Vision Library)
C++
31
star
17

TengineInferPipe

C++
24
star
18

FaceRecognition2

FaceRecognition2 is an implementation project of face detection and recognition. The face detection using MTCNN algorithm, and recognition using LightenedCNN algorithm.
C++
20
star
19

AndroidFacedemo

An face recogniztion detect project on android based on CaffeOnACL
C++
11
star
20

AndroidCVDemo

An CV demo in Android, include Gesture recognition and Object traking
C++
4
star
21

autokernel-docs-en

autokernel documentation (english version)
Python
3
star
22

autokernel-docs-cn

autokernel documentation
Python
1
star