Figment

Detect Hands

Detect one or more hands in the input image and draw them as a skeleton on a solid background. The node runs the MediaPipe hand detector and hand landmark models on the GPU through ONNX Runtime; the MediaPipe library itself is not used. Only the landmark coordinates come back, 21 per hand.

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.
  • Coloring solid draws every point and line in the colors below. per hand draws a right hand in red and a left hand in cyan. per finger gives each finger its own hue as well, warm colors for the right hand and cool colors for the left, so no two fingers of either hand share a color. Use per hand or per finger when the drawing is the input of an image-to-image model such as pix2pix. The point and line colors are ignored in these modes. Handedness is judged from an un-mirrored image: a mirrored webcam swaps the colors.
  • Draw Points Draw each landmark as a dot.
  • Color (points) The color of the dots.
  • Radius The size of the dots.
  • Draw Lines Connect the landmarks of each hand with lines.
  • Color (lines) The color of the lines.
  • Line Width The thickness of the lines.
  • Number of Hands The maximum number of hands to detect, from 1 to 4. Each extra hand costs one more landmark pass.
  • Confidence The minimum detection score for a hand to count, from 0 to 1.
  • Smoothing Filters the landmarks over time 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 is a good starting point. The filter follows each hand from frame to frame by position, so it works without tracking. The drawn skeleton and the Landmarks output are both smoothed. Keep it at 0 for unrelated images, such as a Load Image Folder.

Outputs

  • Out The drawing on the background color, at the size of the input image.
  • Detected true while at least one hand is in the frame.
  • Landmarks An object with type: 'hand' and three arrays with one entry per hand: landmarks (points with x, y, z normalized to the image), handedness (whether the hand is a left or a right hand, with a score), and worldLandmarks (the same points in meters, relative to the hand’s center). The output is null when no hand is detected.