Anchor Box
The short answer · 9 words
In short
Predefined reference box for predicting object positions in images.
Flow-alerta.cab (2026): Anchor Box. https://flow-alerta.cab/en/glossar/anchor-box. Editorially reviewed on: 09 October 2026.
How much detail would you like?
Practical
Explains a building block of anchor-based image models and possible causes of inappropriate recognitions.
Example scenario: A detector adjusts a rectangular reference box to the position of a package in the camera image.
Check appropriate settings for own image projects for model architecture and object sizes.
At a glance
- Core
- Anchor boxes are reference rectangles with defined sizes and aspect ratios at positions of a feature maps. An object detector predicts, among other things, displacements and size corrections. The anchor box is therefore not yet a recognized object boundary.
- Use
- Explains a building block of anchor-based image models and possible causes of inappropriate recognitions.
- Example
- Example scenario: A detector adjusts a rectangular reference box to the position of a package in the camera image.
- Limits
- Confuse reference boxes with finished detections; choose unsuitable sizes; assume that all detectors use anchor boxes.
Questions about this term
Does every object recognition need anchor boxes?
No, there are also anchor-free procedures.
Is every box a real object?
No, predictions have to be evaluated.
All details and background
Anchor boxes are reference rectangles with defined sizes and aspect ratios at positions of a feature maps. An object detector predicts, among other things, displacements and size corrections. The anchor box is therefore not yet a recognized object boundary.
ELI5 – Explained simply
You put stencils on a picture. The model moves and modifies them so that they match possible objects.
Why it matters in practice
Explains a building block of anchor-based image models and possible causes of inappropriate recognitions.
Example
Example scenario: A detector adjusts a rectangular reference box to the position of a package in the camera image.
Pitfalls and limits
- Confuse reference boxes with finished detections; choose unsuitable sizes; assume that all detectors use anchor boxes.
Origin & history
The Faster-R-CNN work from 2015 describes anchor boxes with different sizes and aspect ratios as references for the Proposal Network region.
Use in mid-sized companies
- Check appropriate settings for own image projects for model architecture and object sizes.
Related terms
- Object recognition
Matching Flow-alerta service
AI opportunity analysis & roadmapAI consulting: AI potential analysis & roadmap on tasks, data and prerequisites of a possible first AI use case.Matching tool / use case
View use casesUse Case: Anchor-based packet detection in a test environment.Sources
- arxiv.orgRen et al., Faster R-CNN (2015)
Quellenprüfung: 09.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.
Editorially reviewed on: