Skip to content

Latest commit

 

History

History
37 lines (26 loc) · 1.44 KB

README.md

File metadata and controls

37 lines (26 loc) · 1.44 KB

Dynjandi Webcam Processing Tool

This tool processes webcam input to allow for effects, mainly background masking so far.

Features

  • CLI to toggle features, show FPS and processing times
    • Method of background identification
    • Simple background masking by comparing the current image with a base image
    • Tensorflow based approach
    • Background replacement
      • Replace background with a green screen for further processing
      • Replace background with a blurred version of the background

How-To (Ubuntu)

  • Install v4l2loopback-dkms.
  • Activate virtual webcam sudo modprobe v4l2loopback video_nr=3 card_label="Dynjandi Output"
  • Install dependencies, python3 venv, and requirements
  • Get base image ./get_base.py (try to remove yourself from the image)
  • Download trained model from here
  • Run main script ./get_diff.py
  • Profit! You can further process the image, e.g. with OBS Studio and make yourself float in space!

Possible future features

  • Better masking
    • A machine learning approach seems to work better ([1] [2]), but I'm not touching NodeJS with a ten-foot pole
  • Improve speed, I currently get about 15 FPS at 1280x720

Naming

Dynjandi is named after the Icelandic waterfall of the same name.

License

The MIT License, copyright (c) 2020 wasmitnetzen