In ontology, a class (or a label) serves as a category that groups together entities sharing similar characteristics and properties. For further explanation of the concept, please read our Term Dictionary.

When classes are set up within an Ontology, they can be assigned to detected objects on annotation tool pages.

• To create a new class, click + Create under the Class category in an Ontology.

• For an existing class card, you can click the Edit button on it to make modifications, or click Open to create its sub-classes.

Please note that you cannot change the Tool Type of an existing class. Sub-classes should maintain the same Tool Type as the parent class.

---

Configure Basic Info

Fill in the basic information for the class on the popup window:

---

Select Tool Type

Tool Types available for each type of dataset may differ. Please refer to the following explanation and select an appropriate one.

🌄 Image

1. Bounding Box

Draw rectangles around objects; commonly used for object detection.

• Size Limit (optional): Set limits for the Width, Height, Area, and Width-to-Height Ratio of the bounding box. A "Class Invalid" error will be displayed if the result fails to meet these limits.

2. Polygon

Outline irregular shapes by connecting multiple points; commonly used for annotating objects with intricate boundaries.

• Number of polygon sides (optional): Specify the number of sides for the polygon (minimum 3) and add a new tool labeled "polygon+number" to the toolbar. When using this tool for drawing, the shape will automatically close once annotators reach the designated number of sides.

• Size Limit (optional): Set limits for the Width, Height, and Area of the polygon. A "Class Invalid" error will be displayed if the result fails to meet these limits.

3. Polyline

Annotate lines in images; commonly used for road tracking or annotating object contours.

• Direction Limit (optional): Set whether the polyline needs to align with the Horizontal or Vertical axes. A "Class Invalid" error will be displayed if the result fails to meet this limit.

4. Keypoint

Mark specific points on objects, such as facial landmarks or joints on the human body.

5. Skeleton

Form a skeletal structure by connecting key points on the human body, typically at joint locations such as elbows, knees, shoulders, and hips; commonly used in human pose estimation.

After selecting Skeleton as the tool type, you need to complete the reference image showing the points to be annotated.

💡Steps to Configure the Skeleton Tool

1. Basics

🖱️Right click and hold to drag the image. Zoom in/out with your mouse scroll wheel. 🖱️Left click to select a point or line.

By default, points and edges have the same color as the class. You can change them individually on the right, or modify the colors of existing points and edges in the left list.

You can also customize the reference image by clicking the Update Picture button.

2. Points

Create: Click the Add button, then 🖱️Left click on the image to place a point.

Edit: Click the Edit button. Select a point on the image or the left list, then hold down 🖱️Left click to drag and adjust its position.

Delete: Click the Delete button, then select a point and press the Backspace key to remove it. Alternatively, you can also delete it from the left list.

3. Edges

Create an Edge by Two Nodes: Click the Connect button, then 🖱️Left click on two points to create an edge. Press the ~ key to toggle directions (from the start to the end node) on or off. You can also click the icon on the left list to create an edge by selecting two nodes.

Create Multiple Edges Sequentially: Click the icon on the left list to select the start and end nodes. It will automatically connect the points between them in sequence.

Create Multiple Edges in a Custom Order: Click the icon on the left list to enter node numbers separated by ",". It will connect the nodes in order automatically.

Edit: Select an edge on the left list, then click the edit icon to modify the start and end nodes.

Delete: Click the Delete button, then select an edge and press the Backspace key to remove it. Alternatively, you can also delete it from the left list.

4. Advanced Configuration

Optionally, you can click the Advanced Configuration button to set up extra details about the tool.

Keypoints Label: Enter a label for each point (displayed as a number by default). Each label should be under 255 characters and must not be duplicated.

Equidistant Skeleton: Click the In Order button to set the start and end nodes, or click the Custom button to enter a list of node numbers separated by ",". When annotating in Tools, the points will be evenly distributed along the drawn curve or polyline based on the settings.

Skeleton Attribute Configuration: In addition to class attributes, you can also set attributes for each point in the skeleton, including Name, Alias, Color, and Hotkey (automatically configured in sequence from 1 to 9 and 0).

6. Curve

Outline objects with complex curved shapes such as road bends, natural contours, and medical imaging vessels.

7. Group

Organize and group related results for improved management and understanding of complex scenes.

8. Mask

Create masks to cover specific areas in images; commonly used for Segmentation.

• Size Limit (optional): Set limits for the Area of the mask. A "Class Invalid" error will be displayed if the result fails to meet these limits.

9. Cuboid

Construct a pseudo-3D cuboid around objects by drawing two rectangles; commonly used for annotating objects with depth information.

10. Circle

Mark circular objects or regions in images.

11. Ellipse

Annotate elliptical shapes in images, useful for objects with oval or elliptical structures.

---

🌌 LiDAR Fusion

1. Cuboid

Draw 3D bounding boxes around objects, typically for annotating the position, size, and orientation.

• Constraints (optional): Set limits for the Length, Width, Height, and Points of the cuboid. An error will be displayed if the result fails to meet these limits.
• Standard (optional): Set the standard Length, Width, Height, and Points of the cuboid used for this class as a reference for annotators.

2. Segmentation

Annotate objects or regions in point clouds by segmenting them into different categories or objects.

• Result Type (required): Specify whether the annotation result should be Instance Segmentation or Semantic Segmentation.

Semantic segmentation involves labeling each point in data with a class label without distinguishing between different instances of the same class.

Instance segmentation goes a step further by not only categorizing each point but also distinguishing between different instances of the same class.

3. Polygon

Draw polygonal areas on a plane; commonly used for annotating object contours or irregular shapes.

4. Polyline

Annotate lines in point clouds; commonly used for representing linear objects like roads, pipelines, etc.

5. Group

Organize and group related results for improved management and understanding of complex scenes.

6. Keypoint

Annotate points on a drawn polyline or polygon.

---

🎥 Audio & Video

1. Clip

Trim and segment media files, allowing for accurate labeling of specific sections or events within the content.

• Length Limit (optional): Set the length limit of the clip. An error will be displayed if the result fails to meet these limits.

---

📄 Text

1. Entity

Annotate text entities in textual data by selecting/double-clicking words or sentences and choosing the class & attribute on the tool page.

2. Relation

Annotate relations between selected text entities by connecting the start entity to the end entity and choosing the class & attribute on the tool page.

• Qualified Entity (required): Select starting/ending entities for workers to mark relations in textual data.

---

🤖 Generative AI

1. Dialog Evaluation

This tool is designed for RLHF Dialogue Evaluation annotaion. On the tool page, hover or click on a conversation bubble and then select the class & attribute for the conversation contents between the User and Bot.

• Target on (required): Choose which is the target of evaluation: User or Bot.

2. Dialog Response

It is designed for SFT Dialogue Response annotation. The configured classes are located at the bottom left of the tool page. Click on one class, and then type in your response in the text box without the need to select any conversation bubble on the data canvas.

• Response As (required): Select to response as User or Bot.
• Constraints (optional): Define the character constraints from Min xx Char to Max xx Char.

---

Manage Attributes

When creating a Class, it is essential to add Attributes (properties) to it. Please refer to the Attributes & Classification page for more details.

---

Map with Model Classes

When running a model to pre-annotate the dataset, classes from the model cannot automatically align with the classes in the dataset if they are named differently. In such cases, manual mapping of the classes is necessary.

Click Model Map in the top right corner of the ontology list. Select the Dataset Classes and the Model Classes (in the second and third columns) you wish to align. The fourth column provides detailed descriptions of the Model Classes for better understanding.

👍

Tips:

Checking the model enables quick selection of all classes in it.

You can only configure one dataset class at a time. If you need to match multiple classes, click confirm and then reopen the panel for further actions.

For example, we can map the dataset class Vehicle with multiple classes in the model Basic LiDAR Object Detection.

After configuration, run the model. The class results from the model will be automatically integrated with the dataset classes.