Annotation Types

This reference documents every annotation type supported by Avala, including their properties and JSON representations.

Overview

Avala supports the following annotation types:

Type	Geometry	Use Cases
Bounding Box	2D rectangle	Object detection, localization
Polygon	2D arbitrary shape	Instance segmentation, precise outlines
3D Cuboid	3D box	Autonomous driving, robotics
Segmentation Mask	Pixel-level mask	Semantic segmentation, panoptic segmentation
Polyline	Open line segments	Lane detection, path annotation
Classification	No geometry	Scene classification, image tagging
Keypoints	Named points with connections	Pose estimation, facial landmarks

All annotation types share a set of common properties in addition to their type-specific fields.

Bounding Box

A 2D axis-aligned rectangular box defined by its top-left corner, width, and height.

Properties

Property	Type	Description
`x`	`float`	X coordinate of the top-left corner (pixels)
`y`	`float`	Y coordinate of the top-left corner (pixels)
`width`	`float`	Width of the box (pixels)
`height`	`float`	Height of the box (pixels)
`label`	`string`	Class label
`attributes`	`object`	Optional key-value attributes

JSON Example

{
  "uid": "ann_01HSN4X9K2M...",
  "type": "bounding_box",
  "label": "car",
  "x": 120.0,
  "y": 85.5,
  "width": 200.0,
  "height": 150.0,
  "attributes": {
    "occlusion": "partial",
    "truncated": false
  },
  "created_by": "user_01H...",
  "created_at": "2025-09-15T10:30:00Z"
}

Polygon

An arbitrary closed polygon defined by an ordered list of vertices.

Properties

Property	Type	Description
`vertices`	`array`	Ordered list of `[x, y]` coordinate pairs (pixels)
`label`	`string`	Class label
`attributes`	`object`	Optional key-value attributes

JSON Example

{
  "uid": "ann_01HSN4X9K2N...",
  "type": "polygon",
  "label": "pedestrian",
  "vertices": [
    [305.0, 120.0],
    [320.0, 115.0],
    [340.0, 130.0],
    [345.0, 200.0],
    [330.0, 210.0],
    [310.0, 205.0],
    [300.0, 150.0]
  ],
  "attributes": {
    "pose": "walking"
  },
  "created_by": "user_01H...",
  "created_at": "2025-09-15T10:32:00Z"
}

3D Cuboid

A 3D bounding box defined by its center position, dimensions (length, width, height), and rotation.

Properties

Property	Type	Description
`center`	`object`	Center position with `x`, `y`, `z` (meters)
`dimensions`	`object`	Size with `length`, `width`, `height` (meters)
`rotation`	`object`	Rotation with `yaw`, `pitch`, `roll` (radians)
`label`	`string`	Class label
`attributes`	`object`	Optional key-value attributes

JSON Example

{
  "uid": "ann_01HSN4X9K2P...",
  "type": "cuboid_3d",
  "label": "vehicle",
  "center": {
    "x": 12.5,
    "y": -3.2,
    "z": 0.8
  },
  "dimensions": {
    "length": 4.5,
    "width": 1.8,
    "height": 1.6
  },
  "rotation": {
    "yaw": 1.57,
    "pitch": 0.0,
    "roll": 0.0
  },
  "attributes": {
    "num_lidar_points": 342,
    "occlusion": "none",
    "activity": "moving"
  },
  "created_by": "user_01H...",
  "created_at": "2025-09-15T10:35:00Z"
}

3D cuboid coordinates use the sensor frame of the point cloud. The yaw value represents rotation around the Z-axis (vertical), which corresponds to the object’s heading direction.

Segmentation Mask

A pixel-level mask that labels every pixel within a region. Masks are stored in run-length encoding (RLE) for efficiency.

Properties

Property	Type	Description
`mask`	`object`	RLE-encoded mask with `counts` and `size`
`label`	`string`	Class label
`instance_id`	`integer`	Instance identifier (for instance segmentation)
`attributes`	`object`	Optional key-value attributes

JSON Example

{
  "uid": "ann_01HSN4X9K2Q...",
  "type": "segmentation",
  "label": "road",
  "mask": {
    "counts": [10, 5, 20, 8, 15, 3, 40],
    "size": [1080, 1920]
  },
  "instance_id": 1,
  "attributes": {
    "surface": "asphalt"
  },
  "created_by": "user_01H...",
  "created_at": "2025-09-15T10:38:00Z"
}

Mask Format

The mask field uses COCO-style run-length encoding:

counts: Alternating runs of background and foreground pixel counts
size: [height, width] of the image

Polyline

An open line defined by an ordered sequence of points. Unlike polygons, polylines are not closed.

Properties

Property	Type	Description
`points`	`array`	Ordered list of `[x, y]` coordinate pairs (pixels)
`label`	`string`	Class label
`attributes`	`object`	Optional key-value attributes

JSON Example

{
  "uid": "ann_01HSN4X9K2R...",
  "type": "polyline",
  "label": "lane_marking",
  "points": [
    [400.0, 600.0],
    [420.0, 500.0],
    [450.0, 400.0],
    [490.0, 300.0],
    [540.0, 200.0]
  ],
  "attributes": {
    "line_type": "dashed",
    "color": "white"
  },
  "created_by": "user_01H...",
  "created_at": "2025-09-15T10:40:00Z"
}

Classification

An image-level or frame-level label with no spatial geometry. Used for categorizing entire images or frames.

Properties

Property	Type	Description
`label`	`string`	Classification label
`confidence`	`float`	Confidence score (0.0 to 1.0), typically set by auto-labeling models
`attributes`	`object`	Optional key-value attributes

JSON Example

{
  "uid": "ann_01HSN4X9K2S...",
  "type": "classification",
  "label": "highway",
  "confidence": 0.95,
  "attributes": {
    "weather": "clear",
    "time_of_day": "daytime"
  },
  "created_by": "user_01H...",
  "created_at": "2025-09-15T10:42:00Z"
}

Keypoints

A set of named points with connections between them, typically used for pose estimation and skeleton annotation.

Properties

Property	Type	Description
`keypoints`	`array`	List of keypoint objects, each with `name`, `x`, `y`, and `visibility`
`connections`	`array`	List of `[name_a, name_b]` pairs defining skeleton edges
`label`	`string`	Class label
`attributes`	`object`	Optional key-value attributes

Keypoint Visibility Values

Value	Meaning
`0`	Not present (not in frame or not applicable)
`1`	Occluded (estimated position, not directly visible)
`2`	Visible

JSON Example

{
  "uid": "ann_01HSN4X9K2T...",
  "type": "keypoints",
  "label": "person",
  "keypoints": [
    { "name": "nose", "x": 250.0, "y": 100.0, "visibility": 2 },
    { "name": "left_eye", "x": 245.0, "y": 95.0, "visibility": 2 },
    { "name": "right_eye", "x": 255.0, "y": 95.0, "visibility": 2 },
    { "name": "left_shoulder", "x": 230.0, "y": 140.0, "visibility": 2 },
    { "name": "right_shoulder", "x": 270.0, "y": 140.0, "visibility": 2 },
    { "name": "left_elbow", "x": 215.0, "y": 180.0, "visibility": 2 },
    { "name": "right_elbow", "x": 285.0, "y": 180.0, "visibility": 1 },
    { "name": "left_wrist", "x": 200.0, "y": 210.0, "visibility": 2 },
    { "name": "right_wrist", "x": 295.0, "y": 215.0, "visibility": 0 }
  ],
  "connections": [
    ["left_shoulder", "right_shoulder"],
    ["left_shoulder", "left_elbow"],
    ["left_elbow", "left_wrist"],
    ["right_shoulder", "right_elbow"],
    ["right_elbow", "right_wrist"]
  ],
  "attributes": {
    "action": "standing"
  },
  "created_by": "user_01H...",
  "created_at": "2025-09-15T10:45:00Z"
}

Common Properties

Every annotation in Avala includes these shared properties regardless of type.

Property	Type	Description
`uid`	`string`	Unique annotation identifier
`type`	`string`	Annotation type (e.g., `bounding_box`, `polygon`, `cuboid_3d`)
`label`	`string`	Class label from the project taxonomy
`attributes`	`object`	Optional key-value pairs for additional metadata (e.g., occlusion, truncation)
`tracking_id`	`string`	Object tracking identifier for video/sequence annotations (same across frames)
`frame_index`	`integer`	Frame number within the sequence (for temporal data)
`created_by`	`string`	User ID of the annotator who created the annotation
`created_at`	`string`	ISO 8601 timestamp of creation
`updated_by`	`string`	User ID of the last person to modify the annotation
`updated_at`	`string`	ISO 8601 timestamp of the last modification

The tracking_id field is only present for annotations in video or multi-frame sequence data. It links the same object across frames for consistent tracking.

Next Steps

Learn how to use these tools in the Annotation Tools guide
See Supported Formats for input data formats
Explore Keyboard Shortcuts for efficient annotation

Mission Control

Tutorials

Guides

Reference

Overview

Bounding Box

Properties

JSON Example

Polygon

Properties

JSON Example

3D Cuboid

Properties

JSON Example

Segmentation Mask

Properties

JSON Example

Mask Format

Polyline

Properties

JSON Example

Classification

Properties

JSON Example

Keypoints

Properties

Keypoint Visibility Values

JSON Example

Common Properties

Next Steps

Mission Control

Tutorials

Guides

Reference

​Overview

​Bounding Box

​Properties

​JSON Example

​Polygon

​Properties

​JSON Example

​3D Cuboid

​Properties

​JSON Example

​Segmentation Mask

​Properties

​JSON Example

​Mask Format

​Polyline

​Properties

​JSON Example

​Classification

​Properties

​JSON Example

​Keypoints

​Properties

​Keypoint Visibility Values

​JSON Example

​Common Properties

​Next Steps

Overview

Bounding Box

Properties

JSON Example

Polygon

Properties

JSON Example

3D Cuboid

Properties

JSON Example

Segmentation Mask

Properties

JSON Example

Mask Format

Polyline

Properties

JSON Example

Classification

Properties

JSON Example

Keypoints

Properties

Keypoint Visibility Values

JSON Example

Common Properties

Next Steps