Implementation:Roboflow Rf detr RFDETR Export

Knowledge Sources	RF-DETR RF-DETR Export
Domains	Deployment, Model_Export
Last Updated	2026-02-08 15:00 GMT

Overview

Concrete tool for exporting RF-DETR models to ONNX format provided by the RF-DETR library.

Description

RFDETR.export() delegates to Model.export() which prepares the model, generates sample input tensors via make_infer_image(), and calls export_onnx() to invoke torch.onnx.export with appropriate input/output names and opset version. The export supports backbone-only export, custom input shapes, and optional post-export simplification.

Usage

Call model.export() on any initialized RFDETR model to produce an ONNX file.

Code Reference

Source Location

Repository: rf-detr
File: rfdetr/detr.py
Lines: L129-135 (RFDETR.export)
File: rfdetr/main.py
Lines: L554-624 (Model.export)
File: rfdetr/deploy/export.py
Lines: L64-86 (export_onnx)

Signature

class RFDETR:
    def export(self, **kwargs) -> None:
        """Export your model to an ONNX file."""

class Model:
    def export(
        self,
        output_dir: str = "output",
        infer_dir: str = None,
        simplify: bool = False,
        backbone_only: bool = False,
        opset_version: int = 17,
        verbose: bool = True,
        force: bool = False,
        shape: Optional[Tuple[int, int]] = None,
        batch_size: int = 1,
        **kwargs,
    ) -> None:
        """Export the trained model to ONNX format."""

def export_onnx(
    output_dir: str,
    model: nn.Module,
    input_names: List[str],
    input_tensors: torch.Tensor,
    output_names: List[str],
    dynamic_axes: Optional[Dict],
    backbone_only: bool = False,
    verbose: bool = True,
    opset_version: int = 17,
) -> str:
    """Export model to ONNX using torch.onnx.export. Returns output file path."""

Import

from rfdetr import RFDETRBase
# Internal: from rfdetr.deploy.export import export_onnx

I/O Contract

Inputs

Name	Type	Required	Description
output_dir	str	No	Directory for ONNX file (default: "output")
opset_version	int	No	ONNX opset version (default: 17)
simplify	bool	No	Run simplification after export (default: False)
backbone_only	bool	No	Export only the backbone (default: False)
shape	Optional[Tuple[int,int]]	No	Input shape; defaults to model resolution
batch_size	int	No	Batch size for export (default: 1)

Outputs

Name	Type	Description
ONNX file	File	{output_dir}/inference_model.onnx with inputs ['input'] and outputs ['dets', 'labels']

Usage Examples

Basic Export

from rfdetr import RFDETRBase

model = RFDETRBase()
model.export(output_dir="./onnx_output")
# Produces: ./onnx_output/inference_model.onnx

Export with Simplification

from rfdetr import RFDETRBase

model = RFDETRBase(pretrain_weights="output/checkpoint_best_total.pth")
model.export(
    output_dir="./onnx_output",
    simplify=True,
    opset_version=17,
)
# Produces: ./onnx_output/inference_model.onnx
#           ./onnx_output/inference_model.sim.onnx

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment