Export Configuration

Fine-tune how your exported CSV is structured by customizing column names, delimiters, empty values, and per-label value mappings.

Overview

The export configuration panel appears when you click Export in the dataset view. All settings apply to the current export and are remembered for future exports within the same project. The configuration options vary slightly depending on whether you are using Single Column or Multi Column mode.

Column Name Customization

You can control the name of the label column(s) in the exported CSV using a pattern string with token placeholders. Tokens are enclosed in curly braces and are replaced with actual values at export time.

Available Tokens

Token	Description	Example Value
`{name}`	The display name of the label	`Reach`
`{id}`	The unique identifier of the label	`lbl_abc123`
`{exportValue}`	The configured export value (slug) of the label	`reach`

The column name pattern is primarily used in Multi Column mode where each label becomes its own column. In Single Column mode, you set a single column name directly (no tokens needed).

Column Name Pattern Examples(text)

Pattern: label_{exportValue}
Result:  label_reach, label_grasp, label_lift

Pattern: {name}_annotation
Result:  Reach_annotation, Grasp_annotation, Lift_annotation

Pattern: col_{id}
Result:  col_lbl_abc123, col_lbl_def456, col_lbl_ghi789

Default pattern

The default column name pattern is label_{name}, which produces column names based on the label display name like label_reach.

Delimiter Selection

In Single Column mode, when a frame has multiple active labels, they are joined into a single string using the delimiter you specify. Common delimiter choices include:

Delimiter	Example Output	Notes
`,` (comma)	`"reach,grasp"`	Default. Requires quoting since CSV uses commas as field separators.
`\|` (pipe)	`reach\|grasp`	Good alternative. Rarely appears in label names.
`;` (semicolon)	`reach;grasp`	Good alternative if pipes are used elsewhere.
(space)	`reach grasp`	Works well for NLP tokenization pipelines.

Comma delimiter

If you choose a comma as your label delimiter, the exported values will be wrapped in quotes to avoid breaking the CSV structure. This may require additional parsing in some tools.

Empty Value Specification

The empty value determines what appears in the CSV for frames that have no labels assigned. This applies to both layout modes:

Single Column: The empty value replaces the entire cell content for unlabeled frames.
Multi Column: The empty value is used as the negative value (typically 0) for all label columns on unlabeled frames.

Common empty value choices:

Value	Use Case
`""` (empty string)	Default. Produces an empty cell in the CSV.
`none`	Explicit marker for no-label state in text pipelines.
`unlabeled`	Human-readable placeholder for review workflows.
`NaN`	Compatible with pandas/NumPy missing value conventions.

Positive / Negative Values

In Multi Column mode, each label column contains either a positive value (label is active) or a negative value (label is inactive) for each frame. You can customize both values:

Positive Value	Negative Value	Use Case
`1`	`0`	Default. Standard binary encoding for ML.
`true`	`false`	Boolean representation for typed systems.
`yes`	`no`	Human-readable for spreadsheet review.
`1.0`	`0.0`	Float representation for numeric pipelines.

Multi Column with Boolean Values(csv)

frame,joint_1,label_reach,label_grasp,label_lift
0,0.12,true,false,false
1,0.15,true,false,false
2,0.20,true,true,false
3,0.25,false,true,false

Per-Label Custom Value Overrides

For advanced use cases, you can override the export value on a per-label basis. This is configured in the Label Management panel when defining your label schema.

Each label has an Export Value field that determines what string is used in the CSV output. By default, this is a lowercase slug of the label name (e.g., "Reach" becomes reach), but you can set it to any custom string:

Numeric codes: Set export values to 0, 1, 2, etc. for classification indices.
Custom slugs: Use domain-specific identifiers like action_reach or phase_1.
Abbreviated codes: Use short codes like R, G, L for compact output.

Per-Label Export Values(csv)

frame,joint_1,labels
0,0.12,R
1,0.15,R|G
2,0.20,G
3,0.25,L

Export value vs. display name

The export value is independent of the display name shown in the labeling interface. You can have a friendly display name like "Reach Forward" while exporting as simply reach or 1.

Configuration Examples

Here are some common configuration combinations for different use cases:

ML Training Pipeline

Configuration(text)

Mode:           Multi Column
Column Pattern: label_{exportValue}
Positive Value: 1
Negative Value: 0

Human Review Spreadsheet

Configuration(text)

Mode:           Single Column
Column Name:    annotations
Delimiter:      ;
Empty Value:    (none)

NLP Tokenization Input

Configuration(text)

Mode:           Single Column
Column Name:    actions
Delimiter:      (space)
Empty Value:    idle