Skip to content

baseapp/DeepSight-Face-Recognition-SDK

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Introduction

Deepsight Face is a binary SDK that runs as an HTTP service. It encapsulates Deep learning models for face detection, gender/age classification, face recognition and provides a REST api for easy inference. Each model is a separate plugin that can be upgraded as new updates are pushed.

deepsight

Since all inference is made offline, there are no limits on the number of API hits.

This repository consists of a collection of example programs written in python demonstrating the capabilities of the SDK.

Getting Started and Installation

Deepsight Face is extremely easy to setup and is available for free.

Deepsight Face is currently supported in Linux and Windows Operating Systems on x64 platform.
It is available with or without GPU Acceleration. The free version comes without it.

Visit this link and download a suitable version for your platform

Windows

  1. Run the setup file and install to a location that will NOT require admin privileges for writing. The default C:\Deepsight_Face is safe.

Install Dependencies

The setup package installs necessary dependencies. However, in case that didn't happen, install these present in the redist folder.

  1. VC++ 2017 Runtime (vc_redist.x64.exe)
  2. Intel MKL BLAS Runtime (c_wproc*.exe)

Install CUDA (GPU Version)

For GPU version of the SDK, you must download and install CUDA

  1. Download CUDA Toolkit 9.0 (Sept 2017) for 64-bit Windows 10 from Nvidia's website
  2. Run the installation. If installation fails, disable Visual Studio Integration in the installer options and try again.
  3. Download cuDNN v7.0.5 (Dec 5, 2017), for CUDA 9.0 for 64-bit Windows 10 from Nvidia's cuDNN Website. If necessary signup for a developer account.
  4. Extract the zip archive to the NVIDIA GPU Computing Toolkit\CUDA\ folder and merge with folders from the archive.

Run

  1. The setup should've created start menu links. Launch Deepsight Face from the start menu link. (or dsFace.exe from the installed folder)
  2. The application should start with a bunch of messages and finally say SERVER READY.
  3. At this point you can start the Demo app using your browser and pointing it to localhost:5000.

Linux

  1. Open a terminal and cd into a directory with non-root access.
  2. Copy the shell installer into this directory.
  3. Run the script using
    chmod +x Deepsight_Face-xxx-Linux.sh
    ./Deepsight_Face-xxx-Linux.sh
  4. Press the space bar to read the EULA and enter y to accept it
  5. Continue through the prompt until extraction is complete.

linux installation

Install Dependencies

Deepsight on Linux requires the OpenBLAS library. It should be available in your distribution repository.

# On Ubuntu
sudo apt-get update
sudo apt-get install libopenblas-dev

Install CUDA (GPU Version)

For GPU version of the SDK, you must download and install CUDA

  1. Download CUDA Toolkit 8.0 GA2 (Feb 2017) for x86_64 Linux from Nvidia's website
  2. Perform the installation.
  3. Download cuDNN v5 (May 27, 2016), for CUDA 8.0 for 64 bit Linux from Nvidia's cuDNN Website. If necessary signup for a developer account.
  4. Extract the zip archive to the /usr/local/cuda-8.0/ folder and merge with folders from the archive.
  5. Make sure to update the PATH with /usr/local/cuda-8.0/bin and LD_LIBRARY_PATH with /usr/local/cuda-8.0/lib64

Run

  1. cd into the directory Deepsight_Face and use ./dsFace to launch the program
  2. The application should start with a bunch of messages and finally say SERVER READY.
  3. At this point you can start the Demo app using your browser and pointing it to localhost:5000.

Usage

  • The application accepts arguments as follows
$ ./dsFace -h
Deepsight Face is a Deep Learning powered face recognition SDK that runs locally as a http service
Usage:
  DeepSight Face [OPTION...]

  -v, --verbose    Print lots of messages; vv increases verbosity
  -b, --benchmark  Run benchmark to evaluate speed
  -k, --key        Prompts license key
  -u, --usage      Print usage stats
  -h, --help       Prints help
  -p, --port arg   specify port at which to serve; default is 5000