Implementation:Roboflow Rf detr ORT InferenceSession

Knowledge Sources	ONNX Runtime API RF-DETR
Domains	Deployment, Inference
Last Updated	2026-02-08 15:00 GMT

Overview

External tool documentation for validating and running inference on exported RF-DETR ONNX models using ONNX Runtime.

Description

onnxruntime.InferenceSession loads an ONNX model and provides a run() method for inference. It automatically selects the best execution provider (CUDA GPU, CPU) and applies runtime graph optimizations. For RF-DETR models, the session expects a single input tensor named input and produces dets (bounding boxes) and labels (class logits) outputs.

Usage

Use after ONNX export to validate model correctness or to run inference in production without PyTorch.

Code Reference

Source Location

• External: onnxruntime Python package

Signature

import onnxruntime as ort

session = ort.InferenceSession(
    onnx_path: str,
    providers: List[str] = ["CUDAExecutionProvider", "CPUExecutionProvider"],
)

results = session.run(
    output_names: Optional[List[str]],
    input_feed: Dict[str, np.ndarray],
)

Import

import onnxruntime as ort

I/O Contract

Inputs

Name	Type	Required	Description
onnx_path	str	Yes	Path to ONNX model file
providers	List[str]	No	Execution providers (default: CUDA + CPU fallback)
input	np.ndarray	Yes	Input image tensor of shape (1, 3, H, W), normalized

Outputs

Name	Type	Description
dets	np.ndarray	Bounding box predictions
labels	np.ndarray	Class logit predictions

Usage Examples

Validate Exported Model

import numpy as np
import onnxruntime as ort

# Load ONNX model
session = ort.InferenceSession(
    "output/inference_model.onnx",
    providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
)

# Create sample input (matches model resolution)
input_tensor = np.random.randn(1, 3, 560, 560).astype(np.float32)

# Run inference
dets, labels = session.run(None, {"input": input_tensor})

print(f"Detections shape: {dets.shape}")  # (1, 300, 4)
print(f"Labels shape: {labels.shape}")    # (1, 300, num_classes)

Related Pages

Implements Principle

• Principle:Roboflow_Rf_detr_ONNX_Runtime_Validation

Requires Environment

• Environment:Roboflow_Rf_detr_ONNX_Export_Environment