Train workflow > Annotate Data

Annotate Data

Purpose

Annotate objects to generate a ground truth data set for the purpose of training and validating detectors using machine learning.

In WormLab , objects are defined with keypoints that represent the locations specific parts of a body or object. Multi-point objects also contain connections between the keypoints. Connections are represented by a source point and an end point. For example, a Termite is composed of three points: Head, Thorax, and Abdomen, and two connections: Head-Thorax and Thorax-Abdomen.

Creating new object definitions, points and connections

  1. Click New then enter a name for the new object definition.

  2. Click Add Point and the point will be added to the Point List table. If desired, rename the point by double clicking on point name in the Point List table and typing in the new name.

  3. Click Add Connection then choose its Start and End points from the dropdown menus in the Point Connectivity table.

Selecting an object definition for annotation

Prior to annotation, select the object type from the Object Definitions dropdown menu.

Annotation

Annotation is the process of generating ground truth data set for training object detectors.

This process differs slightly for single point and multi-points objects:

  • Single point annotation

    To annotate objects defined by a single point such as a Cell or an Egg:

    • Hold the Ctrl key down on the keyboard, scroll the mouse wheel until the circular cursor matches the size of the object, then click on the object in the Image Display window.

      OR

    • Click and drag the mouse in the Image Display window to delineate a bounding box around the object.

  • Multi-point annotation

    Multi-point objects are annotated by clicking on each point location in the order in which they appear in the Point List of its object. For each point, hold the Ctrl key down on the keyboard, scroll the mouse wheel until the circular cursor matches the size of the point, then click on the point in the Image Display window.

Navigation

Training a reliable detector requires multiple frames to be labeled. You can select your own frames by using the video playback controls, or click Next and Previous to navigate between image frames.

There is no set number of annotated frames needed for training a detector. How much annotation is needed is determined via Active Learning.

Edit annotations

Selecting

Click Select in the main ribbon then click on an annotation to select it:

  • Single point: Click anywhere on the lines of the square.

  • Multi-point: Click in the center of one of the circles.

Editing

  • Reposition: (Single point annotations only) Hover over the annotation until the cursor changes into four arrows, then click and drag the annotation to the new position in the Image Display window.

  • Re-size: (Single point annotations only) Hover over a square at the edge of the annotation until the cursor changes into two arrows, then click and drag to move that edge of the annotation.

  • Delete: Press Delete on the keyboard.

Once editing is complete, click Start Annotation or click Mark in the main ribbon to continue annotating.

Data Annotation panel descriptions

Object Definitions

  • Object Type: Choose the desired object type from the dropdown menu.

  • New Click to create a new object type. In the Object Type window, type a name into the text box then click OK

  • Delete Click to delete the object type selected in the dropdown menu.

  • Add Point Click to add a new point to the object type selected in the dropdown menu. The new point will appear in the Point List.

  • Add Connection Click to add a new connection associated with the object type selected in the dropdown menu. The new connection will appear in the Point Connectivity list. For each connection, choose the appropriate Start and End point from the dropdown menu.

Annotation

  • Delete Annotations Click to delete existing annotations.

    • This Frame: Choose this option to delete annotations in the current frame only.

    • All Frames: Choose this option to delete all annotations from all frames.

  • Start Annotation Click to start placing annotations on the objects in the image display window.

  • Default Size: Check the box and enter a value to place annotation squares of the specified value.

  • Navigation: Click Previous or Next to navigate between image frames.

Point List

  • Point Name: Lists points associated with the selected object type. For multi-point objects, points are placed in the order in which they appear in the point list.

  • Delete: Click the X to delete a point type.

Point Connectivity: Lists the connections between points.

Point connectivity becomes important with more complex objects. For example, a fly may have a head, thorax, abdomen and two wings. Both wings would be connected to the thorax.

  • Start: From the dropdown menu, choose the start point type.

  • End: From the dropdown menu, choose the end point type.

  • Delete: Click the X to delete a connection.