Skip to content
No description, website, or topics provided.
Python C Other
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
Makefile Update Jul 17, 2019

This repo refers to

General overview

We basically took the pre-trained RetinaFace network from, provided a working Dockerfile, and a python script to process videos.

How to test

NOTE: this has been tester on a ubuntu 16.04 machine, with a Nvidia GPU (GTI 1080), with docker, nvidia-docker installed, and all relevant drivers. We use in Dockerfile mxnet/python:1.4.1_gpu_cu100_py2 as base image. Feel free to go to to use a different tag, depending on your CUDA version.

git clone

Download the .zip file from and put the 2 files (.params and .json) in a folder on your computer (see {{modelPath}} below). Those 2 files are the pre-trained network.

Download input.mp4 from Put it in a folder on your computer (see {{dataPath}} below)

cd fruty_face-detection

docker build -t retinaface .

docker run -it --runtime=nvidia -v {{dataPath}}:/data -v {{modelPath}}:/model retinaface /bin/bash

( replace {{dataPath}} with the local folder on your computer containing input.mp4, replace {{modelPath}} with the local folder on your computer containing the model parameters .params and .json)


In my case it ran for approximately 15min, I was then able to open output.avi, and see the input footage with the human faces highlighted by a green square.

You can’t perform that action at this time.