Skip to main content
The Bounding Box tool draws axis-aligned rectangular boxes around objects. It is the most common annotation type for object detection tasks.

​
When to Use

Use bounding boxes when:
  • You need to localize objects without precise boundary outlines
  • The task requires object detection (e.g., YOLO, Faster R-CNN, SSD)
  • Objects are roughly rectangular or the exact shape is not important
  • Speed is more important than pixel-perfect accuracy
Consider a different tool when:
  • You need precise object boundaries — use the Polygon or Segmentation tool instead
  • You need 3D localization in point clouds — use the 3D Cuboid tool
  • Objects are highly irregular and overlapping — polygons or segmentation masks give better results

​
Usage

  1. Press B or select the Box tool from the toolbar
  2. Click and drag from one corner to the opposite corner
  3. Release to complete the box
  4. Select a label from the label dropdown

​
Editing

  • Move: Click inside the box and drag
  • Resize: Drag any edge or corner handle
  • Adjust precisely: Use the properties panel to enter exact pixel values for x, y, width, and height

​
Shortcuts

ShortcutAction
BActivate the Bounding Box tool
Shift + dragConstrain to a square
Delete / BackspaceDelete selected box
Ctrl+Z / ⌘ZUndo
Ctrl+Shift+Z / ⌘⇧ZRedo

​
Common Mistakes

  • Box too loose: The box should tightly fit the visible extent of the object with minimal padding
  • Missing occluded objects: Include partially occluded objects — the box should cover the visible portion
  • Merging overlapping objects: Draw separate boxes for each distinct object, even when they overlap
  • Including shadows: Unless your project guidelines say otherwise, exclude shadows from the box

​
Advanced Tips

  • Hold Shift while dragging to constrain the box to a perfect square
  • Use the properties panel for pixel-exact adjustments when precision matters
  • In video mode, bounding boxes can be interpolated across frames — annotate key frames and let the system fill in between
  • When objects are partially cut off at the image edge, extend the box to the image boundary
  • Polygon — For precise object outlines with arbitrary shapes
  • 3D Cuboid — For 3D bounding boxes in point clouds
  • Segmentation — For pixel-level object masks
  • Classification — For image-level labels without geometry