Skip to main content

Data Annotation

Data Viewing

Enter the Data page and click on the corresponding file name link to start data viewing and annotation. The data page displays all available datasets, and you can quickly find the data you need through search and filtering functions.

The dataset page displays all available data files

Panels

A Panel is the rendering unit of data, equivalent to a camera perspective (Topic).

Each piece of data typically contains multiple camera perspectives, consisting of multiple panels. You can flexibly adjust the size, position, and number of panels to browse data in the best form. The flexible layout of panels allows you to view data from different angles simultaneously, significantly improving annotation efficiency.

Switching Rendering Panels

You can easily change the display content of any panel:

  1. Click the settings button in the upper left corner of the panel
  2. Select the desired panel type from the dropdown menu

This allows you to quickly switch between different data views according to current task requirements.

Click the settings button in the upper left corner of the panel

Select the desired panel type from the dropdown menu

Changing Subscription Messages

You can also change the data source displayed by the panel:

  1. Click the settings button in the upper left corner of the panel
  2. Select the Topic message source you want to switch to in the popup settings menu

This allows you to view different data streams to meet the needs of various annotation scenarios.

Change Topic message source through the settings menu

Preview the newly selected Topic content

Fine-grained Annotation

tip

The annotation method in this section is suitable for segmented annotation containing multiple atomic actions, especially suitable for scenarios where one piece of data contains multiple tasks. Fine-grained annotation allows you to mark multiple different actions or events in a single data stream.

For annotation task workflow, please refer to:

Start Annotation

Click the "Start" button (or press shortcut key Q) to mark the start time point of an action or event. The system will record the current timestamp as the starting point of the annotation.

Click the start button or use shortcut key Q to start annotation

You can:

  • Auto play (press spacebar)
  • Manually navigate forward and backward (press left and right arrow keys)
  • Control navigation speed (press CTRL to accelerate or ALT to decelerate)

These flexible playback controls allow you to precisely locate the time points that need annotation.

End Annotation

When the action or event ends, click the "End" button (or press shortcut key R) to end the current annotation. The system will record the current timestamp as the end point of the annotation, thus determining the complete time range.

Click the end button or use shortcut key R to end annotation

Add Semantic Information

After marking the time period, you need to add semantic information to the annotation:

  1. Fill in the form information
  2. Select the corresponding natural language description or add new dictionary items
  3. Save the information

Detailed semantic descriptions help with subsequent data training and analysis.

Fill in the natural language description of the annotation Select predefined descriptions from options or add new descriptions

Add Annotation

After completing the above steps, click the "Add Annotation" button (or press Enter) to add this annotation to the annotation list of the current data. If you need to annotate multiple actions, repeat the above steps.

Click the add annotation button or use Enter key to confirm addition

Save Annotation

After completing all annotations, click the yellow upload button in the upper right corner to save the annotations to the server. Please make sure to save your work before leaving the page to prevent data loss.

Click the save button in the upper right corner to upload annotations to the server

Batch Annotation

tip

This function allows you to apply the annotation results of one piece of data to multiple homogeneous similar data, significantly saving annotation time.

For single task scenarios (such as 100 pieces of data all being the same action), the batch annotation function can greatly improve efficiency.

Batch Annotation Workflow

  1. Select Sample Data: Batch select homogeneous data from the data list Select one piece of data from the list as a sample

  2. Create Annotation Task: Create a new annotation task for this data Create a new annotation task

  3. Start Annotation: View and start executing the annotation task View the created annotation task

  4. Complete Sample Annotation: Complete the annotation of one sample according to the fine-grained annotation process Complete the annotation of sample data

  5. Save Sample: Save the completed sample annotation Save the completed sample annotation

  6. Select Template: Return to the task page and select the annotated sample data as a template Select the sample to use as a template

  7. Batch Copy: Start batch copying annotations to other data Start the batch copy annotation process

  8. Confirm Success: Confirm that the copy operation has been successfully completed System displays successful annotation copying

  9. View Results: View the final results of batch annotation View the final results of batch annotation

Through the batch annotation function, you can quickly apply standardized annotations to large amounts of data, especially suitable for processing large-scale homogeneous datasets, greatly improving work efficiency.