Skip to content

Installation

Requirements

  • Python 3.10+
  • (Optional) CUDA-capable GPU for accelerated inference

uv is the recommended package manager. Install it first, then:

git clone https://github.com/imessam/rf-detr-onnx.git
cd rf-detr-onnx

onnxruntime-gpu is a base dependency — GPU acceleration is enabled by default with uv sync. There is no separate --extra gpu.

# Default install (GPU via onnxruntime-gpu)
uv sync

# With model export support
uv sync --extra export

# With test support
uv sync --extra test

# With documentation tools
uv sync --extra docs

CPU-only override

If you don't have a GPU, override onnxruntime-gpu after syncing:

uv sync
.venv/bin/pip uninstall -y onnxruntime-gpu
.venv/bin/pip install onnxruntime

Using pip

From the repository root:

pip install .

With optional extras:

pip install ".[export]"   # Model export from .pth checkpoints
pip install ".[test]"     # Accuracy testing with pytest
pip install ".[docs]"     # Build documentation locally

CPU-only installation

If you don't have a GPU and onnxruntime-gpu fails to install, replace it with the CPU version:

pip uninstall -y onnxruntime-gpu
pip install onnxruntime

C++ Build Requirements

The C++ implementation requires:

Dependency Version Notes
CMake ≥ 3.15 Build system
OpenCV ≥ 4.x Image processing
ONNX Runtime C++ ≥ 1.17 Inference engine

Build

cd cpp
mkdir build && cd build
cmake .. -DONNXRUNTIME_ROOT_DIR=/path/to/onnxruntime
make -j$(nproc)

Tip

Set ONNXRUNTIME_ROOT_DIR to the root of your ONNX Runtime C++ installation (the folder that contains include/ and lib/). The default search path is /opt/onnxruntime.

Downloading Models

Pre-converted ONNX models are available from:

  • Hugging Face: Download manually.
  • Automatic download: The benchmark script will download models automatically if none are found.
# Manual download helper
bash tools/download_onnx.sh

Place model files in the models/ directory (any subfolder is OK).