开源软件名称（OpenSource Name）：

开源软件地址(OpenSource Url)：

开源编程语言(OpenSource Language)：

开源软件介绍(OpenSource Introduction)：

Linux-Fake-Background-Webcam

Background

Video conferencing software support for background blurring and background replacement under Linux is relatively poor. 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).

This repository started of as a tidy up of Benjamen Elder's blog post. His blogpost described a background replacement solution using Python, OpenCV, Bodypix neural network, which is only available under Tensorflow.js. The scripts in Elder's blogpost do not work out of box. This repository originally provided a turn-key solution for creating a virtual webcam with background replacement and additionally foreground object placement, e.g. a podium.

Over time this repository got strangely popular. However it has been clear over time that Bodypix is slow and difficult to set up. Various users wanted to use their GPU with Tensorflow.js, this does not always work. The extra code that provided GPU support sometimes created problems for CPU-only users.

Recently Google released selfie segmentation support for Mediapipe. This repository has been updated to use Mediapipe for image segmentation. This significantly increased the performance. The older version of this repository is now stored in the bodypix branch.

The performance improvement introduced by 2f7d698 means that you can get at least 25FPS on an i7-4900MQ.

Prerequisite

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. Therefore support for akvcam has been added.

v4l2loopback

The v4l2loopback kernel module can be installed through the package manager of your Linux distribution or compiled from source following the instructions in the v4l2loopback github repository.

Once installed, the module needs to be loaded. This can be done manually for the current session by running

$ sudo modprobe v4l2loopback devices=1 exclusive_caps=1 video_nr=2 card_label="fake-cam"

which will create a virtual video device /dev/video2, however, this will not persist past reboot. (Note that the exclusive_caps=1 option is required for programs such as Zoom and Chrome).

To create the virtual video device on startup, run the v4l2loopback-install.sh script to create /etc/modules-load.d/v4l2loopback.conf to load the module and /etc/modprobe.d/linux-fake-background.conf to configure the module.

The camera will appear as fake-cam in your video source list.

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.

Ubuntu 18.04

If you are using Ubuntu 18.04, and if you want to use v4l2loopback, please compile v4l2loopback from the source. You need to do the following:

1. Remove the v4l2loopback package
   • sudo rmmod -r v4l2loopback
   • sudo apt-get remove v4l2loopback-dkms
2. Install DKMS and the Linux kernel headers
   • sudo apt-get install dkms linux-headers-`uname -r`
3. Install v4l2loopback from the repository
   • git clone https://github.com/umlaeute/v4l2loopback.git
   • cd v4l2loopback
4. Install the module via DKMS
   • sudo cp -R . /usr/src/v4l2loopback-1.1
   • sudo dkms add -m v4l2loopback -v 1.1
   • sudo dkms build -m v4l2loopback -v 1.1
   • sudo dkms install -m v4l2loopback -v 1.1
5. Load the module
   • sudo modprobe v4l2loopback

This may apply for other versions of Ubuntu as well. For more information, please refer to the following Github issue.

Akvcam

To install akvcam, you need to do the following:

1. Install the driver by following the instruction at Akvcam wiki. I recommend installing and managing the driver via DKMS.
2. Configure the driver by copying akvcam to /etc/. Please note that the configuration file I supplied locks your Akvcam instance to a resolution of 1280x720. It is different to the configuration file automatically generated by Webcamoid, as my configuration locks the input/output colour format of Akvcam.
3. Note down the output of ls /dev/video*.
4. Run sudo modprobe akvcam or make the akvcam start with a system: sudo akvcam-install.sh
5. Akvcam should have created two extra video devices.
6. When running fake.py, you need to set -v to the first video device that Akvcam created, e.g. if Akvcam created /dev/video5 and /dev/video6, you need to set -v /dev/video5.
7. The software that uses the virtual webcam should the second device that Akvcam created, e.g. if Akvcam created /dev/video5 and /dev/video6, you need to set the software to use /dev/video6.

Note that in akvcam/config.ini, Akvcam (Output device) is the device that fake.py outputs to, and Akvcam (Capture device) is the "capture device", which is opened by the software that you want to use the virtual webcam with.

You might have to specify the --no-ondemand flag when using Akvcam.

For more information on configuring Akvcam, please refer to Akvcam wiki

Disabling UEFI Secure boot

Both v4l2loopback and Akvcam require custom kernel module. This might not be possible if you have secure boot enabled. Please refer to your device manufacturer's manual on disabling secure boot.

Python 3

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 ${HOME}/.profile. This line is needed for Debian Buster.

export PATH="$HOME/.local/bin":$PATH

Upgrading pip

Mediapipe requires pip version 19.3 or above. (Please refer to here and here). However, the pip distributed with some Linux distributions is outdated, e.g. Debian Buster.

If you are on Debian Buster please make sure .local/bin is in your PATH. You can make sure this is the case by adding:

PATH="$HOME/.local/bin":$PATH

in your ~/.profile.

You can then upgrade pip by running:

pip3 install --upgrade pip

Installation

The actual installation can be done by simply running

./install.sh

Installing with Docker

The use of Docker is no longer supported. I no longer see any reason for using Docker with this software. However I have left behind the files related to Docker, for those who want to fix Docker support. Please also refer to DOCKER.md. The Docker related files were provided by liske.

Docker made starting up and shutting down the virtual webcam more convenient for when Bodypix was needed. The ability to change background and foreground images on-the-fly is unsupported when running under Docker.

Usage

In the terminal window, do the following (if using v4l2loopback) :

python3 fake.py

You configure it using a ini file, see ./config-example.ini. To run with a config file, use the following command:

python3 fake.py -c ./config-example.ini

The files that you might want to replace are the followings:

• background.jpg - the background image
• foreground.jpg - the foreground image
• foreground-mask.jpg - the foreground image mask

By default this program uses on-demand processing. The program processes images from the real webcam only when there are programs using the fake webcam. If there are no programs reading from the fake webcam, this program disconnects the real webcam, pauses processing and outputs a black image at 1 FPS to reduce CPU usage. You can manually toggle between the paused/unpaused state by pressing CTRL-C. Unpausing the program also reload the files listed above. This allows you to replace them without restarting the program. You can disable the on-demand processing behaviour by specifying the --no-ondemand flag.

Note that animated background is supported. You can use any video file that can be read by OpenCV.

fake.py

fakecam.py supports the following options:

usage: fake.py [-h] [-c CONFIG] [-W WIDTH] [-H HEIGHT] [-F FPS] [-C CODEC]
               [-w WEBCAM_PATH] [-v V4L2LOOPBACK_PATH] [--no-background]
               [-b BACKGROUND_IMAGE] [--tile-background] [--background-blur k]
               [--background-blur-sigma-frac frac] [--background-keep-aspect]
               [--no-foreground] [-f FOREGROUND_IMAGE]
               [-m FOREGROUND_MASK_IMAGE] [--hologram] [--no-ondemand]
               [--background-mask-update-speed BACKGROUND_MASK_UPDATE_SPEED]
               [--use-sigmoid] [--threshold THRESHOLD] [--no-postprocess]
               [--select-model SELECT_MODEL] [--cmap-person CMAP_PERSON]
               [--cmap-bg CMAP_BG]

Faking your webcam background under GNU/Linux. Please refer to:
https://github.com/fangfufu/Linux-Fake-Background-Webcam

optional arguments:
  -h, --help            show this help message and exit
  -c CONFIG, --config CONFIG
                        Config file (default: None)
  -W WIDTH, --width WIDTH
                        Set real webcam width (default: 1280)
  -H HEIGHT, --height HEIGHT
                        Set real webcam height (default: 720)
  -F FPS, --fps FPS     Set real webcam FPS (default: 30)
  -C CODEC, --codec CODEC
                        Set real webcam codec (default: MJPG)
  -w WEBCAM_PATH, --webcam-path WEBCAM_PATH
                        Set real webcam path (default: /dev/video0)
  -v V4L2LOOPBACK_PATH, --v4l2loopback-path V4L2LOOPBACK_PATH
                        V4l2loopback device path (default: /dev/video2)
  --no-background       Disable background image and blur the real background
                        (default: False)
  -b BACKGROUND_IMAGE, --background-image BACKGROUND_IMAGE
                        Background image path, animated background is
                        supported. (default: background.jpg)
  --tile-background     Tile the background image (default: False)
  --background-blur k   The gaussian bluring kernel size in pixels (default:
                        21)
  --background-blur-sigma-frac frac
                        The fraction of the kernel size to use for the sigma
                        value (ie. sigma = k / frac) (default: 3)
  --background-keep-aspect
                        Crop background if needed to maintain aspect ratio
                        (default: False)
  --no-foreground       Disable foreground image (default: False)
  -f FOREGROUND_IMAGE, --foreground-image FOREGROUND_IMAGE
                        Foreground image path (default: foreground.jpg)
  -m FOREGROUND_MASK_IMAGE, --foreground-mask-image FOREGROUND_MASK_IMAGE
                        Foreground mask image path (default: foreground-
                        mask.png)
  --hologram            Add a hologram effect (default: False)
  --no-ondemand         Continue processing when there is no application using
                        the virtual webcam (default: False)
  --background-mask-update-speed BACKGROUND_MASK_UPDATE_SPEED
                        The running average percentage for background mask
                        updates (default: 50)
  --use-sigmoid         Force the mask to follow a sigmoid distribution
                        (default: False)
  --threshold THRESHOLD
                        The minimum percentage threshold for accepting a pixel
                        as foreground (default: 75)
  --no-postprocess      Disable postprocessing (masking dilation and blurring)
                        (default: True)
  --select-model SELECT_MODEL
                        Select the model for MediaPipe. For more information,
                        please refer to https://github.com/fangfufu/Linux-
                        Fake-Background-
                        Webcam/issues/135#issuecomment-883361294 (default: 1)
  --cmap-person CMAP_PERSON
                        Apply colour map to the person using cmapy. For
                        examples, please refer to
                        https://gitlab.com/cvejarano-
                        oss/cmapy/blob/master/docs/colorize_all_examples.md
                        (default: None)
  --cmap-bg CMAP_BG     Apply colour map to background using cmapy (default:
                        None)

Args that start with '--' (eg. -W) can also be set in a config file (specified
via -c). Config file syntax allows: key=value, flag=true, stuff=[a,b,c] (for
details, see syntax at https://goo.gl/R74nmi). If an arg is specified in more
than one place, then commandline values override config file values which
override defaults.

License

The source code of this repository are released under GPLv3.

Linux Fake Background Webcam
Copyright (C) 2020-2021  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/>.