Skip to main content
The Classification tool applies labels at the image level or frame level, without drawing any geometry. Use it for scene-level metadata like weather, lighting, or scene type.

​
When to Use

Use classification when:
  • You need to label the entire image or frame, not specific objects
  • The task involves scene classification (weather, time of day, scene type)
  • You need metadata tags for filtering or stratifying datasets
  • You want to combine scene-level context with object-level annotations from other tools
Consider a different tool when:

​
Usage

  1. Press K or select the Classification tool from the toolbar
  2. The classification panel appears (or the label panel switches to classification mode)
  3. Select one or more labels that describe the image/frame:
    • Weather: sunny, rainy, foggy
    • Scene: highway, urban, parking lot
    • Time of day: daytime, nighttime
  4. Classifications are saved automatically when you navigate to the next item

​
Multi-Label vs. Single-Label

  • Single-label: Only one classification can be selected (radio buttons)
  • Multi-label: Multiple classifications can be selected (checkboxes)
This is configured at the project level when defining the label taxonomy. Check your project setup to understand which mode applies.

​
Shortcuts

ShortcutAction
KActivate the Classification tool
Ctrl+Z / ⌘ZUndo
Ctrl+Shift+Z / ⌘⇧ZRedo

​
Common Mistakes

  • Applying object-level labels at the scene level: Classification is for the entire image, not for individual objects — use bounding boxes or polygons for object-level labeling
  • Inconsistent labeling criteria: Define clear rules for ambiguous cases (e.g., is a cloudy sky with patches of blue “sunny” or “cloudy”?)
  • Forgetting to classify: Unlike drawing annotations, classification has no visible geometry to remind you — make sure every image has its classification labels set

​
Advanced Tips

  • Use classification for metadata that applies to the entire image, not to specific objects
  • Combine with object-level annotations: classification for scene context, boxes/polygons for objects
  • In video sequences, classification can vary per frame (e.g., lighting changes during a clip)
  • Classification labels are useful for building dataset subsets — filter by scene type to create focused training splits