Skip to content

Manually Labeling Multimodal Data Sets

Manual labeling in dFL allows users to directly annotate complex, multimodal datasets through an intuitive interface. This feature is designed for workflows where precise, context-aware input from the user is essential, such as exploratory analysis, creating benchmark datasets, or refining automated pipelines.

Workflow Overview

1. Selecting Records

  • Data sets are organized into records, selectable from the Select Record drop-down menu.
  • A record can represent, e.g.:
    1. A specific shot or discharge number
    2. A simulation model type
    3. A diagnostic component
    4. Any other user-defined data grouping

Once selected, all available signals for that record are loaded into dFL.

2. Visualizing Signals

  • Add signals to the workspace using Add Graph.

  • Supported visualization types are fully interactive, and discussed in the Supported graphs section.

3. Creating Labels

  • Define a new label in the Settings tab. Note that labels must be separated by spaces (and thus may not contain spaces). Multi-word labels should use delimiters such as underscores, e.g. "Multi_word_labels". Labels must be saved after creating by hitting the "Save Settings" button on the Settings tab.

  • The label then appears in the Select Labels drop-down under the Main Control Panel options.
  • Label structures, categories, or metadata can also be pre-defined in the data provider ingestion script.

4. Applying Labels Manually

Two manual labeling methods are available:

  1. Direct Entry
    • Select the label under "Select Labels" in the main Control Panel options (note that mutiple labels may be slected at once). Then manually input the start and end times for the label in the control panel.

  1. Interactive Selection
    • Enable Label Selection Mode from the Main Control Panel..
    • Highlight the region of interest directly on the graph.
    • Confirm with Confirm Label to commit the label.

All active labels appear in the Current Labels tab under the Main Control Panel.

5. Update Labels Manually

Labels may be easily updated after they have been created, using either:

  1. Direct Entry
    • Select the label of interest by clicking on it.
    • Select "Update Existing Labels" in the main Control Panel options, and manually input the updated start and end times for the label in the control panel.
    • Click "Update Selected" Label in the Main Control Panel options.

  1. Interactive Selection
    • Make sure "Label Selection Mode" is turned off under Main Control Panel options.
    • Select the label of interest by clicking on it.
    • Move or resize the label as required.
    • Click "Update Selected" Label in the Main Control Panel options.

6. Exporting Labels

  • Labels can be exported as csv files directly from the Current Labels tab.
    • You may export the entire label set, or manualy select a subset of labels for export.
  • Exported files are ready for:
    1. Downstream analysis
    2. Archiving
    3. Integration into machine learning workflows
    4. etc.