• Home
  • History
  • Annotate
Name Date Size #Lines LOC

..03-May-2022-

.github/H25-May-2021-6,0255,751

benchmark/H03-May-2022-4,6064,414

cmake/H25-May-2021-1,108934

docs/H25-May-2021-5,9944,579

examples/H03-May-2022-5,5744,073

glslang/H25-May-2021-

images/H03-May-2022-

python/H03-May-2022-8,2635,970

src/H03-May-2022-382,075302,389

tests/H03-May-2022-13,57110,325

toolchains/H25-May-2021-1,2521,039

tools/H03-May-2022-46,22136,467

.astylercH A D25-May-2021623 3629

.clang-formatH A D25-May-20213.7 KiB133130

.gitattributesH A D25-May-202130 21

.gitignoreH A D25-May-2021406 4938

.gitmodulesH A D25-May-2021185 76

.restyled.yamlH A D25-May-2021339 2316

.travis.ymlH A D25-May-20211.4 KiB4840

CONTRIBUTING.mdH A D25-May-2021414 64

Info.plistH A D25-May-2021575 1918

README.mdH A D25-May-202117.3 KiB182131

build-android.cmdH A D25-May-20212.9 KiB6051

build.shH A D25-May-20216.6 KiB220164

codeformat.shH A D25-May-20211,018 148

package.shH A D25-May-20216.9 KiB172151

pyproject.tomlH A D25-May-2021162 98

setup.pyH A D25-May-20216 KiB162108

README.md

1![](https://raw.githubusercontent.com/Tencent/ncnn/master/images/256-ncnn.png)
2# ncnn
3
4[![License](https://img.shields.io/badge/license-BSD--3--Clause-blue.svg)](https://raw.githubusercontent.com/Tencent/ncnn/master/LICENSE.txt)
5[![Build Status](https://travis-ci.org/Tencent/ncnn.svg?branch=master)](https://travis-ci.org/Tencent/ncnn)
6[![download](https://img.shields.io/github/downloads/Tencent/ncnn/total.svg)](https://github.com/Tencent/ncnn/releases)
7[![codecov](https://codecov.io/gh/Tencent/ncnn/branch/master/graph/badge.svg)](https://codecov.io/gh/Tencent/ncnn)
8[![Language grade: C/C++](https://img.shields.io/lgtm/grade/cpp/g/Tencent/ncnn.svg?logo=lgtm&logoWidth=18)](https://lgtm.com/projects/g/Tencent/ncnn/context:cpp)
9
10ncnn is a high-performance neural network inference computing framework optimized for mobile platforms. ncnn is deeply considerate about deployment and uses on mobile phones from the beginning of design. ncnn does not have third party dependencies. it is cross-platform, and runs faster than all known open source frameworks on mobile phone cpu. Developers can easily deploy deep learning algorithm models to the mobile platform by using efficient ncnn implementation, create intelligent APPs, and bring the artificial intelligence to your fingertips. ncnn is currently being used in many Tencent applications, such as QQ, Qzone, WeChat, Pitu and so on.
11
12ncnn 是一个为手机端极致优化的高性能神经网络前向计算框架。ncnn 从设计之初深刻考虑手机端的部署和使用。无第三方依赖，跨平台，手机端 cpu 的速度快于目前所有已知的开源框架。基于 ncnn，开发者能够将深度学习算法轻松移植到手机端高效执行，开发出人工智能 APP，将 AI 带到你的指尖。ncnn 目前已在腾讯多款应用中使用，如 QQ，Qzone，微信，天天P图等。
13
14---
15
16### 技术交流QQ群：637093648(超多大佬)  答案：卷卷卷卷卷
17
18### Telegram Group https://t.me/ncnnyes
19
20### Discord Channel https://discord.gg/YRsxgmF
21
22---
23
24### Current building status matrix
25
26| System | CPU (32bit) | CPU (64bit) | GPU (32bit) | GPU (64bit) |
27| :---: | :---: | :---: | :--: | :--: |
28| Linux (GCC) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-x86-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-x86-cpu-gcc) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-x64-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-x64-cpu-gcc) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-x64-gpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-x64-gpu-gcc) |
29| Linux (Clang) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-x86-cpu-clang)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-x86-cpu-clang) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-x64-cpu-clang)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-x64-cpu-clang) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-x64-gpu-clang)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-x64-gpu-clang) |
30| Linux (ARM) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-arm-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-arm-cpu-gcc) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-aarch64-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-aarch64-cpu-gcc) | — | — |
31| Linux (MIPS) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-mips-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-mips-cpu-gcc) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-mips64-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-mips64-cpu-gcc) | — | — |
32| Linux (RISC-V) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/linux-riscv64-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Alinux-riscv64-cpu-gcc) | — | — |
33| Windows (VS2015) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x86-cpu-vs2015)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x86-cpu-vs2015) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x64-cpu-vs2015)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x64-cpu-vs2015) | — | — |
34| Windows (VS2017) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x86-cpu-vs2017)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x86-cpu-vs2017) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x64-cpu-vs2017)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x64-cpu-vs2017) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x64-gpu-vs2017)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x64-gpu-vs2017) |
35| Windows (VS2019) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x86-cpu-vs2019)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x86-cpu-vs2019) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x64-cpu-vs2019)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x64-cpu-vs2019) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/windows-x64-gpu-vs2019)](https://github.com/Tencent/ncnn/actions?query=workflow%3Awindows-x64-gpu-vs2019) |
36| MacOS | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/macos-x64-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Amacos-x64-cpu) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/macos-x64-gpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Amacos-x64-gpu) |
37| MacOS (ARM) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/macos-arm64-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Amacos-arm64-cpu) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/macos-arm64-gpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Amacos-arm64-gpu) |
38| Android | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-armv7-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-armv7-cpu) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-armv8-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-armv8-cpu) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-armv7-gpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-armv7-gpu) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-armv8-gpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-armv8-gpu) |
39| Android-x86 | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-x86-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-x86-cpu) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-x64-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-x64-cpu) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-x86-gpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-x86-gpu) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/android-x64-gpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aandroid-x64-gpu) |
40| iOS | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/ios-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aios-cpu) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/ios-cpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aios-cpu) | — | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/ios-arm64-gpu)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aios-arm64-gpu) |
41| iOS Simulator | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/ios-simulator)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aios-simulator) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/ios-simulator)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aios-simulator) | — | — |
42| WebAssembly | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/web-assembly)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aweb-assembly) | — | — | — |
43| RISC-V GCC/Newlib | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/elf-riscv32-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aelf-riscv32-cpu-gcc) | [![Build Status](https://img.shields.io/github/workflow/status/Tencent/ncnn/elf-riscv64-cpu-gcc)](https://github.com/Tencent/ncnn/actions?query=workflow%3Aelf-riscv64-cpu-gcc) | — | — |
44
45---
46
47### Support most commonly used CNN network
48### 支持大部分常用的 CNN 网络
49
50* Classical CNN: [VGG](https://github.com/BVLC/caffe/wiki/Model-Zoo#models-used-by-the-vgg-team-in-ilsvrc-2014) [AlexNet](https://github.com/BVLC/caffe/tree/9b891540183ddc834a02b2bd81b31afae71b2153/models/bvlc_alexnet) [GoogleNet](https://github.com/BVLC/caffe/tree/9b891540183ddc834a02b2bd81b31afae71b2153/models/bvlc_googlenet) Inception ...
51* Practical CNN: [ResNet](https://github.com/tornadomeet/ResNet) [DenseNet](https://github.com/liuzhuang13/DenseNet) [SENet](https://github.com/hujie-frank/SENet) [FPN](https://github.com/unsky/FPN) ...
52* Light-weight CNN: [SqueezeNet](https://github.com/forresti/SqueezeNet) [MobileNetV1](https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet_v1.md)/[V2/V3](https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/README.md) [ShuffleNetV1](https://github.com/farmingyard/ShuffleNet)/[V2](https://github.com/opconty/keras-shufflenetV2) [MNasNet](https://github.com/tensorflow/models/tree/master/research/slim/nets/nasnet) ...
53* Face Detection: [MTCNN](https://github.com/ipazc/mtcnn) [RetinaFace](https://github.com/biubug6/Pytorch_Retinaface) ...
54* Detection: [VGG-SSD](https://github.com/lzx1413/CAFFE_SSD) [MobileNet-SSD](https://github.com/chuanqi305/MobileNet-SSD) [SqueezeNet-SSD](https://github.com/chuanqi305/SqueezeNet-SSD) [MobileNetV2-SSDLite](https://github.com/chuanqi305/MobileNetv2-SSDLite) [MobileNetV3-SSDLite](https://github.com/XiaoyuHuang96/MobilenetV3SSDLite-tfkeras) ...
55* Detection: [Faster-RCNN](https://github.com/rbgirshick/py-faster-rcnn) [R-FCN](https://github.com/daijifeng001/R-FCN) ...
56* Detection: [YOLOV2](https://github.com/longcw/yolo2-pytorch) [YOLOV3](https://github.com/ultralytics/yolov3) [MobileNet-YOLOV3](https://github.com/eric612/MobileNet-YOLO) [YOLOV4](https://github.com/Tianxiaomo/pytorch-YOLOv4) [YOLOV5](https://github.com/ultralytics/yolov5) ...
57* Detection: [NanoDet](https://github.com/RangiLyu/nanodet)
58* Segmentation: [FCN](https://github.com/unsky/FPN) [PSPNet](https://github.com/hszhao/PSPNet) [UNet](https://github.com/zhixuhao/unet) [YOLACT](https://github.com/dbolya/yolact) ...
59* Pose Estimation: [SimplePose](https://github.com/dog-qiuqiu/Ultralight-SimplePose) ...
60
61---
62
63### HowTo
64
65**[how to build ncnn library](https://github.com/Tencent/ncnn/wiki/how-to-build) on Linux / Windows / MacOS / Raspberry Pi3 / Android / NVIDIA Jetson / iOS / WebAssembly**
66
67* [Build for Linux / NVIDIA Jetson / Raspberry Pi3](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-linux)
68* [Build for Windows x64 using VS2017](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-windows-x64-using-visual-studio-community-2017)
69* [Build for MacOS](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-macos)
70* [Build for ARM Cortex-A family with cross-compiling](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-arm-cortex-a-family-with-cross-compiling)
71* [Build for Hisilicon platform with cross-compiling](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-hisilicon-platform-with-cross-compiling)
72* [Build for Android](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-android)
73* [Build for iOS on MacOS with xcode](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-ios-on-macos-with-xcode)
74* [Build for WebAssembly](https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-webassembly)
75
76**[download prebuild binary package for android and ios](https://github.com/Tencent/ncnn/releases)**
77
78**[use ncnn with alexnet](https://github.com/Tencent/ncnn/wiki/use-ncnn-with-alexnet) with detailed steps, recommended for beginners :)**
79
80**[ncnn 组件使用指北 alexnet](https://github.com/Tencent/ncnn/wiki/use-ncnn-with-alexnet.zh) 附带详细步骤，新人强烈推荐 :)**
81
82**[use netron for ncnn model visualization](https://netron.app)**
83
84**[out-of-the-box web model conversion](https://convertmodel.com/#outputFormat=ncnn)**
85
86[ncnn low-level operation api](https://github.com/Tencent/ncnn/wiki/low-level-operation-api)
87
88[ncnn param and model file spec](https://github.com/Tencent/ncnn/wiki/param-and-model-file-structure)
89
90[ncnn operation param weight table](https://github.com/Tencent/ncnn/wiki/operation-param-weight-table)
91
92[how to implement custom layer step by step](https://github.com/Tencent/ncnn/wiki/how-to-implement-custom-layer-step-by-step)
93
94---
95
96### FAQ
97
98**[ncnn throw error](https://github.com/Tencent/ncnn/wiki/FAQ-ncnn-throw-error)**
99
100**[ncnn produce wrong result](https://github.com/Tencent/ncnn/wiki/FAQ-ncnn-produce-wrong-result)**
101
102**[ncnn vulkan](https://github.com/Tencent/ncnn/wiki/FAQ-ncnn-vulkan)**
103
104---
105
106### Features
107
108* Supports convolutional neural networks, supports multiple input and multi-branch structure, can calculate part of the branch
109* No third-party library dependencies, does not rely on BLAS / NNPACK or any other computing framework
110* Pure C++ implementation, cross-platform, supports android, ios and so on
111* ARM NEON assembly level of careful optimization, calculation speed is extremely high
112* Sophisticated memory management and data structure design, very low memory footprint
113* Supports multi-core parallel computing acceleration, ARM big.LITTLE cpu scheduling optimization
114* Supports GPU acceleration via the next-generation low-overhead vulkan api
115* The overall library size is less than 700K, and can be easily reduced to less than 300K
116* Extensible model design, supports 8bit quantization and half-precision floating point storage, can import caffe/pytorch/mxnet/onnx/darknet/keras/tensorflow(mlir) models
117* Support direct memory zero copy reference load network model
118* Can be registered with custom layer implementation and extended
119* Well, it is strong, not afraid of being stuffed with 卷   QvQ
120
121### 功能概述
122
123* 支持卷积神经网络，支持多输入和多分支结构，可计算部分分支
124* 无任何第三方库依赖，不依赖 BLAS/NNPACK 等计算框架
125* 纯 C++ 实现，跨平台，支持 android ios 等
126* ARM NEON 汇编级良心优化，计算速度极快
127* 精细的内存管理和数据结构设计，内存占用极低
128* 支持多核并行计算加速，ARM big.LITTLE cpu 调度优化
129* 支持基于全新低消耗的 vulkan api GPU 加速
130* 整体库体积小于 700K，并可轻松精简到小于 300K
131* 可扩展的模型设计，支持 8bit 量化和半精度浮点存储，可导入 caffe/pytorch/mxnet/onnx/darknet/keras/tensorflow(mlir) 模型
132* 支持直接内存零拷贝引用加载网络模型
133* 可注册自定义层实现并扩展
134* 恩，很强就是了，不怕被塞卷 QvQ
135
136---
137
138### supported platform matrix
139
140* ✅ = known work and runs fast with good optimization
141* ✔️ = known work, but speed may not be fast enough
142* ❔ = shall work, not confirmed
143* / = not applied
144
145|    |Windows|Linux|Android|MacOS|iOS|
146|---|---|---|---|---|---|
147|intel-cpu|✔️|✔️|❔|✔️|/|
148|intel-gpu|✔️|✔️|❔|❔|/|
149|amd-cpu|✔️|✔️|❔|✔️|/|
150|amd-gpu|✔️|✔️|❔|❔|/|
151|nvidia-gpu|✔️|✔️|❔|❔|/|
152|qcom-cpu|❔|✔️|✅|/|/|
153|qcom-gpu|❔|✔️|✔️|/|/|
154|arm-cpu|❔|❔|✅|/|/|
155|arm-gpu|❔|❔|✔️|/|/|
156|apple-cpu|/|/|/|✔️|✅|
157|apple-gpu|/|/|/|✔️|✔️|
158
159
160---
161
162### Example project
163
164* https://github.com/nihui/ncnn-android-squeezenet
165* https://github.com/nihui/ncnn-android-styletransfer
166* https://github.com/nihui/ncnn-android-mobilenetssd
167* https://github.com/moli232777144/mtcnn_ncnn
168* https://github.com/nihui/ncnn-android-yolov5
169* https://github.com/nihui/ncnn-android-scrfd ��
170
171<img src="https://github.com/nihui/ncnn-assets/raw/master/20181217/ncnn-2.jpg" width="360" height="640"/> <img src="https://github.com/nihui/ncnn-assets/raw/master/20181217/4.jpg" width="360" height="640"/>
172<img src="https://github.com/nihui/ncnn-assets/raw/master/20181217/ncnn-33.jpg" width="360" height="640"/> <img src="https://github.com/nihui/ncnn-assets/raw/master/20181217/ncnn-m.png" width="360" height="640"/>
173<img src="https://github.com/nihui/ncnn-android-yolov5/raw/master/screenshot.jpg" width="360" height="800"/> <img src="https://github.com/nihui/ncnn-android-scrfd/raw/master/screenshot.jpg" width="360" height="800"/>
174
175
176---
177
178### License
179
180[BSD 3 Clause](LICENSE.txt)
181
182