mediapipe face detection Cog model

This is an implementation of the mediapipe's face detection as a Cog model. Cog packages machine learning models as standard containers

It allows batch or individual face detection, and outputs a mask of the face position(s)

Basic usage

    cog predict \
  -i images=@path/to/file \
  -i blur_amount=1.5 \
  -i bias=0 \
  -i output_transparent_image=true

or without cloning the git repo:

cog predict r8.im/chigozienri/mediapipe-face@latest \
  -i images=@path/to/file \
  -i blur_amount=0.0 \
  -i bias=10 \
  -i output_transparent_image=false

• images (required) can be a path to a png/jpg/jpeg, or zip/tar of multiple png/jpg/jpeg
• blur_amount is any float >= 0, the amount of blur applied to the edges of the mask
• bias is an int between 0-255, how light the background of the mask should be (if you want to let some of the original background through)
• output_transparent_image is a bool, outputs an RGBA of the original image with the mask on the alpha channel if true, or a grayscale image of the mask if false