Video conferencing software support for background blurring and background replacement under Linux is relatively poor. The Linux version of Zoom only supports background replacement via chroma key. The Linux version of Microsoft Team does not support background blur. Over at Webcamoid, we tried to figure out if we can do these reliably using open source software (issues/250).
Benjamen Elder wrote a blog post, describing a background replacement solution using Python, OpenCV, Tensorflow and Node.js. The scripts in Elder's blogpost do not work out of box. In this repository, I tidied up his scripts, and provide a turn-key solution for creating a virtual webcam with background replacement and additionally foreground object placement, e.g. a podium.
Unlike the original blog post this can work with CPU-only. It checks for the presence
of /dev/nvidia0 to determine if there is a GPU present. By
downscaling the image sent to bodypix neural network, and upscaling the
received mask, this whole setup runs sufficiently fast under Intel i7-4900MQ.
You can either use this project with docker or install everything on your host.
If you want to use docker, Please refer to DOCKER.md. The updated Docker related files were
added by liske.
Using Docker is unnecessary. However it makes starting up and shutting down
the virtual webcam very easy and convenient.
You need to install either v4l2loopback or akvcam. This repository was originally written with v4l2loopback in mind. However, there has been report that v4l2loopback does not work with certain versions of Ubuntu. Additionally, the author has never really managed to get v4l2loopback to work with Google Chrome. Therefore support for akvcam has been added.
If you are on Debian Buster, you can do the following:
sudo apt install v4l2loopback-dkms
I added module options for v4l2loopback by creating
/etc/modprobe.d/v4l2loopback.conf with the following content:
options v4l2loopback devices=1 exclusive_caps=1 video_nr=2 card_label="v4l2loopback"
exclusive_caps is required by some programs, e.g. Zoom and Chrome.
video_nr specifies which /dev/video* file is the v4l2loopback device.
In this repository, I assume that /dev/video2 is the virtual webcam, and
/dev/video0 is the physical webcam.
I also created /etc/modules-load.d/v4l2loopback.conf with the following content:
v4l2loopback
This automatically loads v4l2loopback module at boot, with the specified module options.
If you get an error like
OSError: [Errno 22] Invalid argument
when opening the webcam from Python, please try the latest version of v4l2loopback from the its Github repository, as the version from your package manager may be too old.
If you are using Ubuntu 18.04, please be aware of the following additional instruction:
- remove apt package
sudo modprobe -r v4l2loopbacksudo apt remove v4l2loopback-dkms
- install aux
sudo apt-get install linux-genericsudo apt install dkms
- install v4l2loopback from the repository
git clone https://github.com/umlaeute/v4l2loopback.gitcd v4l2loopbackmake# The other commands are not needed
- instal mod
sudo cp -R . /usr/src/v4l2loopback-1.1sudo dkms add -m v4l2loopback -v 1.1sudo dkms build -m v4l2loopback -v 1.1sudo dkms install -m v4l2loopback -v 1.1
- reboot
sudo reboot
This may apply for other versions of Ubuntu as well. For more information, please refer to the following Github issue.
To install akvcam, you need to do the following:
- Install the driver by following the instruction at Akvcam wiki. I recommend installing and managing the driver via DKMS.
- Configure the driver by copying
fakecam/akvcamto/etc/, for more information, please refer to Akvcam wiki
The configuration file I supplied was originally generated by Webcamoid, I
added the rw attributes to do the virtual camera devices. If you already
have already configured akvcam via webcamoid, you need to modify the
/etc/akvcam/config.ini to add the rw attributes.
You will need Python 3. You need to have pip installed. Please make sure that
you have installed the correct version pip, if you have both Python 2 and
Python 3 installed. Please make sure that the command pip3 runs.
In Debian, you can run
sudo apt-get install python3-pip
I am assuming that you have set up your user environment properly, and when you install Python packages, they will be installed locally within your home directory.
You might want to add the following line in your .profile. This line is
needed for Debian Buster.
export PATH="$HOME/.local/bin":$PATH
configuration files yourself.
You need to have Node.js. Node.js version 12 is known to work. To install Node.js, please follow the instructions at NodeSource.
Simply run
./install.sh
To find the device number of your webcam you might consider v4l2-ctl:
sudo apt-get install v4l-utils
v4l2-ctl --list-devices
Assuming you are not using the Docker version, please also make sure that your
TCP port 127.0.0.1:9000 is free, as we will be using it.
You can change the port by setting the environment variable PORT. If you set a path, it will use a UNIX Socket instead.
You need to open two terminal windows. In one terminal window, do the following:
cd bodypix
node app.js
In the other terminal window, do the following:
cd fakecam
python3 fake.py
The files that you might want to replace are the followings:
fakecam/background.jpg- the background imagefakecam/foreground.jpg- the foreground imagefakecam/foreground-mask.jpg- the foreground image mask
If you want to change the files above in the middle of streaming, replace them
and press CTRL-C
Note that animated background is supported. You can use any video file that can be read by OpenCV.
If you are not running fakecam.py under Docker, it supports the following options:
usage: fake.py [-h] [-W WIDTH] [-H HEIGHT] [-F FPS] [-S SCALE_FACTOR]
[-B BODYPIX_URL] [-w WEBCAM_PATH] [-v V4L2LOOPBACK_PATH]
[--akvcam] [-i IMAGE_FOLDER] [-b BACKGROUND_IMAGE]
[--tile-background] [--no-foreground] [-f FOREGROUND_IMAGE]
[-m FOREGROUND_MASK_IMAGE] [--hologram]
Faking your webcam background under GNU/Linux. Please make sure your bodypix
network is running. For more information, please refer to:
https://github.com/fangfufu/Linux-Fake-Background-Webcam
optional arguments:
-h, --help show this help message and exit
-W WIDTH, --width WIDTH
Set real webcam width
-H HEIGHT, --height HEIGHT
Set real webcam height
-F FPS, --fps FPS Set real webcam FPS
-S SCALE_FACTOR, --scale-factor SCALE_FACTOR
Scale factor of the image sent to BodyPix network
-B BODYPIX_URL, --bodypix-url BODYPIX_URL
Tensorflow BodyPix URL (or path to UNIX socket)
-w WEBCAM_PATH, --webcam-path WEBCAM_PATH
Set real webcam path
-v V4L2LOOPBACK_PATH, --v4l2loopback-path V4L2LOOPBACK_PATH
V4l2loopback device path
--akvcam Use an akvcam device rather than a v4l2loopback device
-i IMAGE_FOLDER, --image-folder IMAGE_FOLDER
Folder which contains foreground and background images
-b BACKGROUND_IMAGE, --background-image BACKGROUND_IMAGE
Background image path, animated background is
supported.
--tile-background Tile the background image
--no-foreground Disable foreground image
-f FOREGROUND_IMAGE, --foreground-image FOREGROUND_IMAGE
Foreground image path
-m FOREGROUND_MASK_IMAGE, --foreground-mask-image FOREGROUND_MASK_IMAGE
Foreground mask image path
--hologram Add a hologram effect
If under/over-segmentation occurs, you can tweak segmentationThreshold. To
make the network run faster, you can change internalResolution, however this
will reduce segmentation accuracy.
Tensorflow.js uses Tensorflow C library. The default version shipped with
Tensorflow.js is most likely not optimised for your CPU. For me, it gives me
warnings that it was not compiled AVX2 and FMA instructions. Compiling
a version of Tensorflow C library that is optimised for your CPU will improve
the performance. For me, it improved the framerate.
In order to compile your own Tensorflow C library, please follow the instruction at TENSORFLOW.md
Linux Fake Background Webcam
Copyright (C) 2020 Fufu Fang
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <https://www.gnu.org/licenses/>.
Please note that Benjamen Elder's blog post is licensed under CC BY 4.0 (see the bottom of that webpage). According to FSF, CC BY 4.0 is a noncopyleft license that is compatible with the GNU General Public License version 3.0 (GPLv3), meaning I can adapt a CC BY 4.0 licensed work, forming a larger work, then release it under the terms of GPLv3.