Comment on page

Feature Points

Label for Key Point Group Detection


Feature Points (also known as interest points or key points) enable key point detection for a group.
Architectures such as PoseNet and OpenPose support feature points as input. Common use cases include facial recognition and body pose detection.
Key points of the face defined

Define a Feature Points Label

Plainsight does not support model training for this label type.
  1. 1.
    Navigate to the "Label Definitions" tab of your dataset
  2. 2.
    Type label name in the Label Name field
  3. 3.
    Choose Feature Points from the Label Type dropdown
  4. 4.
    Select label color from the color chooser menu
  5. 5.
    Select Start Color and End Color from the chooser menu. The group of points will begin and end with a set of colors in that spectrum.
  6. 6.
    Enter the number of points in the Count field
  7. 7.
    Click "Save" button
  8. 8.
    Repeat as needed for multiple labels
After you've defined a Feature Point label, you can start using it to label a group of points on an image.

Data format

A Feature Point schema element creates an array of objects with data consisting of an array of nullable points. Each points is a XY coordinate relative (0.0 to 1.0) to width and height respectively.
"type": "featurePoints",
"data": {
"points": [
[0.42994670357189385, 0.5286462454172425],
[0.28166641810948584, 0.630226165611773],
[0.2443887262097693, 0.5045251725296982],
[0.3858563286899797, 0.2287688113601787], null
"children": {}

Labeling Feature Points

Once you've defined your feature point label, you can start annotating your data.

Train a Model with Feature Points

Plainsight doesn't currently support SmartML model training with feature points. Feature point annotations can be exported using COCO and trained outside of Plainsight.
Last modified 1yr ago