When to Use
Use keypoints when:- The task requires pose estimation (human, animal, or custom skeletons)
- You need to annotate body landmarks (joints, facial features, hand points)
- The downstream model expects keypoint annotations (e.g., OpenPose, HRNet, MediaPipe)
- You need to track articulated objects across video frames
- You need object localization without pose — use the Bounding Box tool
- You need pixel-level masks — use the Segmentation tool
- You need to outline an object shape — use the Polygon tool
Usage
- Press
Jor select the Keypoint tool from the toolbar - A skeleton template appears based on your project’s keypoint definition
- Click to place the first keypoint
- Place subsequent keypoints following the skeleton structure
- The viewer draws connections between keypoints automatically based on the skeleton definition
- Assign a label
Skeleton Definitions
Skeleton definitions are configured at the project level. A skeleton specifies:- Keypoint names: e.g., nose, left_eye, right_eye, left_shoulder
- Connections: which keypoints are connected by edges
- Visibility: each keypoint can be marked as visible, occluded, or not present
Editing Keypoints
- Move a keypoint: Click and drag it
- Set visibility: Right-click a keypoint to toggle between visible, occluded, and not present
- Move entire skeleton: Select all keypoints and drag
Shortcuts
| Shortcut | Action |
|---|---|
J | Activate the Keypoint tool |
Delete / Backspace | Delete selected skeleton |
Ctrl+Z / ⌘Z | Undo |
Ctrl+Shift+Z / ⌘⇧Z | Redo |
Visibility States
Each keypoint has one of three visibility states:| State | Meaning | When to use |
|---|---|---|
| Visible | The keypoint is clearly visible in the image | Default for most keypoints |
| Occluded | The keypoint exists but is hidden behind another object | You can estimate its location from context |
| Not present | The keypoint is not in the image at all | The body part is outside the frame or does not exist |
Common Mistakes
- Guessing occluded locations: Mark keypoints as “not present” rather than guessing their location if they are fully hidden with no contextual clues
- Wrong skeleton assignment: Make sure the keypoint you are placing matches the correct body part in the skeleton definition
- Inconsistent occluded vs. not present: Use occluded when you can reasonably estimate the location (e.g., hand behind back), not present when you cannot
- Placing on clothing instead of body: Keypoints should be placed on the anatomical joint location, not on the surface of clothing
Advanced Tips
- Place the most visible keypoints first, then fill in occluded ones
- Use zoom for precise placement on small or distant subjects
- Keypoints can be tracked across video frames for pose tracking workflows
- For crowd scenes, place each person’s skeleton separately, even when they overlap
- Use the camera projection in multi-camera setups to verify 3D keypoint positions
Related Tools
- Bounding Box — For object detection without pose
- Polygon — For precise object outlines
- Segmentation — For pixel-level masks
- Classification — For image-level labels