ncnn is a high-performance neural network forward computing framework optimized for mobile phones. From the beginning of design, ncnn has deeply considered the deployment and use of mobile phone terminals, no third-party dependence, cross-platform, mobile phone cpu speed is faster than all known open source frameworks.
Based on ncnn, developers can easily transplant deep learning algorithms to mobile phones for efficient execution, and develop artificial intelligence apps, which have been used in many Tencent applications, such as QQ, Qzone, wechat, Tiantian Ptu, etc.
ncnn Features Overview:
Support convolutional neural network, support multi-input and multi-branch structure, can calculate some branches
No third-party library dependencies, no BLAS/NNPACK and other computing frameworks
Pure C++ implementation, cross-platform, support android ios, etc
ARM NEON assembler level conscience optimization, extremely fast computing speed
Fine memory management and data structure design, very low memory consumption
Support multi-core parallel computing acceleration, ARM big.LITTLE cpu scheduling optimization
Supports GPU acceleration based on the new low cost vulkan api
The overall library volume is less than 700K and can be easily reduced to less than 300K
Extensible model design, quantitative and half precision floating point support 8 bit storage, can be introduced into caffe/pytorch/mxnet/onnx/darknet/keras/tensorflow (mlir) model
Supports loading network models with direct memory zero copy references
Custom layer implementations can be registered and extended
Platforms supported by nccn:
= Running fast = Speed may not be fast enough = not confirmed / = Not applied
give a typical example
You can read more on your own。
