Figment

Detect Faces

Detect one or more faces in the input image and draw them on a solid background. The node runs the MediaPipe face detector and face landmark models on the GPU through ONNX Runtime; the MediaPipe library itself is not used. The image never leaves the GPU; only the landmark coordinates come back, 478 per face.

Connect the landmarks output to a Send OSC node to send the landmarks to another application.

Parameters

  • Background The color that fills the output image behind the drawing.
  • Draw Mode What to draw for each face. contours draws the outline of the face, eyes, eyebrows, and lips. tesselation draws the full face mesh. bounding box draws a rectangle around the landmarks.
  • Color The drawing color.
  • Line Width The thickness of the lines.
  • Number of Faces The maximum number of faces to detect, from 1 to 4. Each extra face costs one more landmark pass, so keep this at the number you need.
  • Confidence The minimum detection score for a face to count, from 0 to 1. Raise it to reject false detections, lower it to keep faces at difficult angles.
  • Mode video follows each face from frame to frame and only runs the detector again when a face is lost. This is faster and steadier for a webcam or movie. still runs the detector on every frame. Use it for unrelated images, such as a Load Image Folder.
  • Smoothing Filters the landmarks over time in video mode so they stop jittering. 0 is off. Higher values are steadier but add a little lag on fast movement; the filter follows fast motion more closely than slow motion, so around 0.65 (the strength MediaPipe uses) is a good starting point. The drawing and the Landmarks output are both smoothed. Has no effect in still mode.

Outputs

  • Out The drawing on the background color, at the size of the input image.
  • Detected true while at least one face is in the frame.
  • Landmarks An object with type: 'face' and a landmarks array with one entry per face. Each entry is a list of points with x, y, and z normalized to the image (0 to 1). The output is null when no face is detected.

Example

The PIX2PIX tutorial uses this node with the tesselation draw mode to turn photos of faces into training data.