walkerlau / gpu-cnn

Accelerate convolution neural network for face recognition using GPU

Accelerate CNN computation with GPU

Please indicate the source below when this repo is shared somewhere else:

Source code: https://github.com/WalkerLau/GPU-CNN

The purpose of this repo is to accelerate VIPLFaceNet, a face recognition algorithm with 7 convolution layers, with GPU and CUDA programming.

For more infomation about VIPLFaceNet:

https://github.com/seetaface/SeetaFaceEngine

Development Environment

• HARDWARE

  • Machine: Amazon AWS EC2, g3s.xlarge (Ohio, us-east-2)

  • GPU: Tesla M60

  • Compute Capability: 5.2

• SOFTWARE

  • AMI ID: Deep Learning Base AMI (Ubuntu) Version 19.1 (ami-04fde1417a0168c95)

  • Ubuntu 16.04 (contained in AMI)

  • OpenCV 2.4.9.1 (contained in AMI)

  • CUDA 10.0.130 (contained in AMI)

Installation

1. Download this repo

   git clone https://github.com/WalkerLau/GPU-CNN.git

   Note: The model file GPU-CNN/model/seeta_fr_v1.0.bin is stored in Git Large File Storage (LFS). You should Download and install Git Large File Storage (LFS) before further operations on Linux machine, or download the seeta_fr_v1.0.bin independently from this repo's webpage. Simply "git clone" would not download the complete seeta_fr_v1.0.bin and will cause error. The complete seeta_fr_v1.0.bin should be of about 110 MB.

2. Build

   cd GPU-CNN
   
   mkdir build
   
   cd build
   
   cmake ../src
   
   make
   

   Note: Try switching to root if build failed.

3. Run

   ./GPU-CNN

Others

• About source files:

  • test_face_recognizer.cpp contains top level function

  • conv_net.cpp calls convolution layers

  • inner_product_net.cpp calls fully-connected layers

  • math_functions.cu defines CUDA kernel functions

References

• Best Practices Guide

• Programming Guide

• CUDA API References

gpu-cnn's Issues

Git clone

您好，我是从西电睿思上面看到一直关注之前FPGA的那个神经网络的加速项目，现在想要下载您的这个GPU的这个项目，但是里面的modul里面的那个大文件一直下载不下来，请问有别的方式下载您的整个工程文件吗？