Skip to content

RF-DETR

Real-time object detection sized for edge devices

Visit official site

What RF-DETR is

The model is described as a transformer-based real-time object detector developed by Roboflow for edge devices. It is stated to exceed 60 AP on the Microsoft COCO benchmark and to keep a compact size for low-latency deployment.

What you can do with it

  • Run accurate object detection at a high frame rate in dynamic scenes
  • Fine-tune a COCO pre-trained model for a specific application
  • Deploy the compact model on edge devices for fast response

Application scenarios

  • Detect and track suspicious people in a security monitoring system
  • Identify and classify objects on the road in an autonomous car
  • Help a robot recognise obstacles in its environment

Getting started

  • Install Python and the required libraries such as requests and PIL
  • Load the pre-trained model with the RFDETRBase class
  • Load the image to be detected into memory
  • Call the predict method of the model for object detection

Pros & cons

✓ What we like

  • Accurate detection at a high frame rate is described
  • Ready-to-run code and examples are provided for integration

! What to watch out for

  • The compact design targets deployment on edge devices
  • Fine-tuning builds on a COCO pre-trained model

FAQ

What kind of model is it?

A transformer-based real-time object detector developed by Roboflow.

What benchmark result is stated?

The material states over 60 AP on the Microsoft COCO benchmark and names RF100-VL.

How is a prediction run?

Install Python with requests and PIL, load RFDETRBase, prepare an image and call predict.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review