Skip to content
    Flow-alerta.cab

    Detection Transformer (DETR)

    AI fundamentals & modelsDETROLDetection Transformer
    Editorially reviewed on: 1 source
    Your reading view

    The short answer · 10 words

    In short

    A Transformer-based model for direct end-to-end object recognition in images.

    Flow-alerta.cab (2026): Detection Transformer (DETR). https://flow-alerta.cab/en/glossar/detection-transformer-detr. Editorially reviewed on: 09 October 2026.

    How much detail would you like?

    Practical

    Relevant as an architecture for object recognition; test suitability for an application based on own images and runtime measurements.

    Possible example: A team tests DETR for the recognition and localization of certain components on conveyor belt images.

    Quality control in production, automated inventory in the warehouse and monitoring of security areas.

    At a glance

    Core
    DETR considers object recognition as a direct set prediction problem. It eliminates the need for handcrafted components such as Non-Maximum Suppression (NMS) and Anchor Boxes and uses self-attention for global image contexts.
    Use
    Relevant as an architecture for object recognition; test suitability for an application based on own images and runtime measurements.
    Example
    Possible example: A team tests DETR for the recognition and localization of certain components on conveyor belt images.
    Limits
    The original DETR required comparatively long training times and had difficulties with small objects; newer variants have to be evaluated separately.

    Questions about this term

    What is a feature of the original DETR?

    It uses volume prediction and requires neither Anchor Boxes nor the usual downstream non-maximum suppression.

    Is it basically better or faster than other detectors?

    No. Accuracy and runtime depend on variant, data and hardware.

    All details and background

    DETR considers object recognition as a direct set prediction problem. It eliminates the need for handcrafted components such as Non-Maximum Suppression (NMS) and Anchor Boxes and uses self-attention for global image contexts.

    ELI5 – Explained simply

    The model predicts a lot of detected objects and their image positions. Whether it recognizes an object correctly must be checked on the basis of test images.

    Why it matters in practice

    Relevant as an architecture for object recognition; test suitability for an application based on own images and runtime measurements.

    Example

    Possible example: A team tests DETR for the recognition and localization of certain components on conveyor belt images.

    Pitfalls and limits

    • The original DETR required comparatively long training times and had difficulties with small objects; newer variants have to be evaluated separately.

    Origin & history

    2020 by Carion et al. (Facebook AI Research) in 'End-to-End Object Detection with Transformers'.

    Use in mid-sized companies

    • Quality control in production, automated inventory in the warehouse and monitoring of security areas.

    Related terms

    • Transformers
    • Computer vision
    • Object recognition
    • YOLO
    • Boundary box

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI Potential Analysis & Roadmap – Evaluating Opportunities for Image Recognition

    Matching tool / use case

    View use casesUse Case: Evaluate object recognition with DETR on operational test images.

    Sources

    • arxiv.orgCarion et al. (2020), End-to-End Object Detection with Transformers

    Quellen geprüft am 09.10.2026.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases