RF-DETR

Use Roboflow's RF-DETR model through our Serverless Hosted API

RF-DETR is Roboflow's transformer-based real-time object detection and segmentation model. You can run inference against COCO-pretrained RF-DETR checkpoints through the Serverless Hosted API, or self-host using Roboflow Inference.

Code samples

The samples below run RF-DETR through the Serverless Hosted API and visualize results with supervision. Pass Roboflow's API Key via the API_KEY env variable.

Object detection

pip install inference-sdk supervision

import os
import urllib.request

import cv2
import supervision as sv
from inference_sdk import InferenceHTTPClient

image_url = "https://storage.googleapis.com/com-roboflow-marketing/notebooks/examples/cars-highway.png"
image_path = "cars-highway.png"
urllib.request.urlretrieve(image_url, image_path)

client = InferenceHTTPClient(
    api_url="https://serverless.roboflow.com",
    api_key=os.getenv("API_KEY"),
)
result = client.infer(image_path, model_id="rfdetr-base")

image = cv2.imread(image_path)
detections = sv.Detections.from_inference(result)

box_annotator = sv.BoxAnnotator()
label_annotator = sv.LabelAnnotator()

annotated = box_annotator.annotate(scene=image.copy(), detections=detections)
annotated = label_annotator.annotate(scene=annotated, detections=detections)

cv2.imwrite("cars-highway-annotated.png", annotated)

Instance segmentation

pip install inference-sdk supervision

import os
import urllib.request

import cv2
import supervision as sv
from inference_sdk import InferenceHTTPClient

image_url = "https://storage.googleapis.com/com-roboflow-marketing/notebooks/examples/bicycle.png"
image_path = "bicycle.png"
urllib.request.urlretrieve(image_url, image_path)

client = InferenceHTTPClient(
    api_url="https://serverless.roboflow.com",
    api_key=os.getenv("API_KEY"),
)
result = client.infer(image_path, model_id="rfdetr-seg-preview")

image = cv2.imread(image_path)
detections = sv.Detections.from_inference(result)

mask_annotator = sv.MaskAnnotator()
label_annotator = sv.LabelAnnotator()

annotated = mask_annotator.annotate(scene=image.copy(), detections=detections)
annotated = label_annotator.annotate(scene=annotated, detections=detections)

cv2.imwrite("bicycle-annotated.png", annotated)

Default COCO-trained models

Pass any of these aliases as the model_id when running inference. The SDK resolves each alias to the underlying Roboflow project version. Input resolution varies by variant.

Alias	Task	Input Size
rfdetr-base	Object Detection	560x560
rfdetr-nano	Object Detection	384x384
rfdetr-small	Object Detection	512x512
rfdetr-medium	Object Detection	576x576
rfdetr-large	Object Detection	704x704
rfdetr-xlarge	Object Detection	700x700
rfdetr-2xlarge	Object Detection	880x880
rfdetr-seg-preview	Instance Segmentation	560x560
rfdetr-seg-nano	Instance Segmentation	312x312
rfdetr-seg-small	Instance Segmentation	384x384
rfdetr-seg-medium	Instance Segmentation	432x432
rfdetr-seg-large	Instance Segmentation	504x504
rfdetr-seg-xlarge	Instance Segmentation	624x624
rfdetr-seg-2xlarge	Instance Segmentation	768x768

Set api_url to match your deployment target:

• https://serverless.roboflow.com for the Serverless Hosted API.

• http://localhost:9001 for a local Inference server.

• Your Dedicated Deployment URL for a private endpoint.