Kili Technology allows annotating poses with a dedicated interface. The pose estimation annotation consists in detecting the position and orientation of an object.
While labeling, placing a point of the image, will automatically trigger the selection of the next point in the pose list of the selected category. This allows you to annotate faster, saving you the time to select the next point. The different categories, the list of pose points, and their order are defined in the JSON interface. Moreover, when you finish placing the last point of the pose, the first point of the same category will be automatically selected, placing it will begin a new annotation.
You can decide if each point of the pose list is mandatory or optional. A point that is not mandatory can be skipped in the pose estimation with (Shift + right arrow). Skipping a point will automatically select the next one in the pose list.
Setting it up
Pose estimation is a specific job that you can parameterize in the JSON interface.
After having added a pose estimation job, you can define the different categories (There is only one in the image below 'Hand') and then the different points of the category pose (there are five for the category hand in the image below: 'TIP', 'DIP', 'PIP', 'MCP' and 'Wrist'). By default, pose points are not mandatory; you can change them by clicking on the "required" button on the JSON interface form (see the image below) or by setting the "mandatory" property of the point to true in the JSON interface.
Shortcuts are provided to help you annotate quickly :