yolo-rust-ort

A Rust implementation of YOLO using ONNX Runtime (ort)

Overview

This small example demonstrates how to implement Ultralytics YOLOv8/YOLOv10 object detection in Rust using ort crate as a backend for running ONNX models. ort is a wrapper around ONNX Runtime.

Getting Started

Prerequisites

Ensure you have a basic Rust development environment set up. If you want to download a specific YOLO model, you'll also need Python with the ultralytics package installed. Note that it only supports yolov8 and yolov10 models.

Running the demo

1. (Optional) Download the YOLO Model
   To download a specific YOLOv10 model, use the following command:

   python ./download_model.py --model yolov10n.pt

   Alternatively, you can use the pre-included model at ./onnx/yolov10n.onnx or ./onnx/yolov8n.onnx.

2. Run the Rust Application

cargo run -- path/to/your/image.jpg path/to/your/model.onnx model_name use_nms

For example, to run inference of a YOLOv8 model, run the command:

cargo run -- ./sample/people0.jpg  onnx/yolov8n.onnx yolov8 true

NMS is disabled by default. However, you can explicitly enable it by passing a boolean flag. YOLOv10 model doesn't need nms, so you can set it to false for YOLOv10 and true for YOLOv8

3. Check the output The processed image and its corresponding detection results will be saved in the ./output directory. The output will include:

• A JPEG image with bounding boxes drawn with colors according to the class IDs.
• A text file containing the detection results in COCO format.

Requirements

raqote requires some packages installed in the system, and in case of a Linux system, you should have fontconfig (libfontconfig1-dev for Ubuntu, and fontconfig-devel for Fedora) and pkg-config packages installed.

Disclaimer

This project is my first attempt at Rust, so the code can be really messy.

References

ort

ultralytics

THU-MIG