yolo_detector

Pre-launch installation

sudo apt update -y

sudo apt install libopencv-dev pkg-config build-essential cmake libgtk-3-dev libcanberra-gtk3-module llvm-dev libclang-dev clang

Converting the model to onnx

This library uses tipo models.onnx

pip install ultralytics

Download the model you need.

https://huggingface.co/Ultralytics/YOLOv8/tree/main

yolo export model=yolov8m.pt format=onnx opset=12 dynamic=True

You also need to download the class file (coco.names)

https://github.com/pjreddie/darknet/blob/master/data/coco.names

Sample code

Detection

use opencv::{highgui, imgcodecs};
use yolo_detector::YoloDetector;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m.onnx", "coco.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, original_size) = detector.detect(&mat.clone())?;

    let result = detector.draw_detections(mat.clone(), detections, 0.5, 0.5, original_size)?;

    highgui::imshow("YOLOv8 Video", &result)?;
    highgui::wait_key(0)?;

    Ok(())
}

use yolo_detector::YoloDetector;
use opencv::imgcodecs;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m.onnx", "coco.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, original_size) = detector.detect(&mat.clone())?;

    let detections_with_classes =
        detector.get_detections_with_classes(detections, 0.5, 0.5, original_size);

    for (class_name, rect) in detections_with_classes {
        println!("Class: {}, Position: {:?}", class_name, rect);
    }

    Ok(())

//returns values
//Class: person, Position: Rect_ { x: 74, y: 875, width: 41, height: 112 }
//Class: car, Position: Rect_ { x: 184, y: 899, width: 499, height: 141 }
}

Weights

use opencv::{highgui, imgcodecs};
use yolo_detector::YoloDetectorWeights;

fn main() -> opencv::Result<()> {
    let mut detector =
        YoloDetectorWeights::new("yolov4.weights", "yolov4.cfg", "coco.names").unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (class_ids, confidences, boxes) = detector.detect(&mat.clone(), 0.7, 0.4)?;

    let result = detector.draw_detections(&mut mat.clone(), class_ids, confidences, boxes)?;

    highgui::imshow("YOLOv8 Video", &result)?;
    highgui::wait_key(0)?;

    Ok(())
}

OBB

DOTAv1.names

plane
ship
storage tank
baseball diamond
tennis court
basketball court
ground track field
harbor
bridge
large vehicle
small vehicle
helicopter
roundabout
soccer ball field
swimming pool

use opencv::{highgui, imgcodecs};
use yolo_detector::YoloDetector;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m-obb.onnx", "DOTAv1.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, original_size) = detector.detect(&mat.clone())?;

    let result = detector.draw_detections_obb(mat.clone(), detections, 0.5, 0.5, original_size)?;

    highgui::imshow("YOLOv8 Video", &result)?;
    highgui::wait_key(0)?;

    Ok(())
}

use yolo_detector::YoloDetector;
use opencv::imgcodecs;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m-obb.onnx", "DOTAv1.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, original_size) = detector.detect(&mat.clone())?;

    let detections_with_classes =
        detector.get_detections_with_classes_obb(detections, 0.5, 0.5, original_size);

    for (class_name, rect, rotation_angle) in detections_with_classes {
        println!(
            "Class: {}, Position: {:?}, Rotation Angle: {}°",
            class_name, rect, rotation_angle
        );
    }

    Ok(())

//returns values
// Class: ship, Position: Rect_ { x: 110, y: 738, width: 84, height: 25 }, Rotation Angle: 77.65746°
// Class: ship, Position: Rect_ { x: 576, y: 733, width: 65, height: 23 }, Rotation Angle: 56.169453°
}

Classification

ImageNet.names

https://github.com/Elieren/yolo_detector/blob/main/ImageNet.names

use yolo_detector::YoloDetector;
use opencv::imgcodecs;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m-cls.onnx", "ImageNet.names", 224).unwrap();

    let mat = imgcodecs::imread("zebra.jpg", imgcodecs::IMREAD_COLOR)?;

    let result = detector.classify(&mat.clone(), 0.5)?;

    for (class_name, score) in result {
        println!("Class: {}, Score: {}", class_name, score);
    }
    
    Ok(())

//returns values
// Class: zebra, Score: 0.99995613
}

Pose

pose.names

Nose
Left Eye
Right Eye
Left Ear
Right Ear
Left Shoulder
Right Shoulder
Left Elbow
Right Elbow
Left Wrist
Right Wrist
Left Hip
Right Hip
Left Knee
Right Knee
Left Ankle
Right Ankle

use opencv::{highgui, imgcodecs};
use yolo_detector::YoloDetector;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m-pose.onnx", "pose.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, original_size) = detector.detect_pose(&mat.clone())?;

    let result = detector.draw_detections_pose(mat.clone(), detections, 0.5, 0.5, original_size)?;

    highgui::imshow("YOLOv8 Video", &result)?;
    highgui::wait_key(0)?;

    Ok(())
}

use yolo_detector::YoloDetector;
use opencv::imgcodecs;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m-pose.onnx", "pose.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, original_size) = detector.detect_pose(&mat.clone())?;

    let result = detector.get_detections_with_classes_pose(detections, 0.5, 0.5, original_size);

    for (i, keypoints) in result.iter().enumerate() {
        println!("Person {}:", i + 1);
        for (name, point) in keypoints {
            println!("  {}: ({}, {})", name, point.x, point.y);
        }
    }

    Ok(())

//returns values
// Person 1:
//   Nose: (545, 98)
//   Right Shoulder: (524, 115)
//   Left Wrist: (556, 153)
//   Left Hip: (549, 155)
//   Left Shoulder: (552, 115)
//   Left Elbow: (555, 135)
//   Left Eye: (547, 96)
//   Right Eye: (541, 96)
//   Right Wrist: (520, 154)
//   Right Knee: (530, 183)
//   Right Ear: (532, 98)
//   Right Hip: (530, 155)
//   Right Elbow: (516, 136)
// Person 2:
//   Left Hip: (829, 160)
//   Left Knee: (834, 185)
//   Left Shoulder: (820, 122)
//   Right Ear: (851, 102)
//   Left Elbow: (817, 142)
//   Left Ankle: (837, 210)
//   Right Ankle: (847, 210)
//   Left Ear: (824, 103)
//   Right Knee: (849, 185)
//   Right Shoulder: (861, 121)
//   Right Hip: (856, 160)
}

Segmentation

use opencv::{highgui, imgcodecs};
use yolo_detector::YoloDetector;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m-seg.onnx", "coco.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, mask, original_size) = detector.detect_mask(&mat.clone())?;

    let result =
        detector.draw_detections_masked(mat.clone(), detections, mask, 0.5, 0.5, original_size)?;

    highgui::imshow("YOLOv8 Video", &result)?;
    highgui::wait_key(0)?;

    Ok(())
}

use yolo_detector::YoloDetector;
use opencv::imgcodecs;

fn main() -> opencv::Result<()> {
    let detector = YoloDetector::new("yolov8m-seg.onnx", "coco.names", 640).unwrap();

    let mat = imgcodecs::imread("image.jpg", imgcodecs::IMREAD_COLOR)?;

    let (detections, mask, original_size) = detector.detect_mask(&mat.clone())?;

    let detections =
        detector.get_detections_with_classes_masks(detections, mask, 0.5, 0.5, original_size);

    // Обработка результатов
    for (class_name, rect, conf, mask) in detections {
        println!(
            "Class: {}, Confidence: {}, BoundingBox: {:?}, Mask: {:?}",
            class_name, conf, rect, mask
        );
    }

    Ok(())
//returns values
// Class: car, Confidence: 0.92939186, BoundingBox: Rect_ { x: 185, y: 900, width: 500, height: 142 }, Mask: Mat { type: "CV_8UC1", flags: "0x42FF4000 (continuous)", channels: 1, depth: "CV_8U", dims: 2, size: Size_ { width: 1680, height: 1116 }, rows: 1116, cols: 1680, elem_size: 1, elem_size1: 1, total: 1874880, is_continuous: true, is_submatrix: false, data: <element count is higher than threshold: 1000> }
// Class: person, Confidence: 0.7530618, BoundingBox: Rect_ { x: 1091, y: 860, width: 78, height: 133 }, Mask: Mat { type: "CV_8UC1", flags: "0x42FF4000 (continuous)", channels: 1, depth: "CV_8U", dims: 2, size: Size_ { width: 1680, height: 1116 }, rows: 1116, cols: 1680, elem_size: 1, elem_size1: 1, total: 1874880, is_continuous: true, is_submatrix: false, data: <element count is higher than threshold: 1000> }
}

Project roadmap

• Detection
• Weights
• OBB
• Classification
• Pose
• Segmentation

Note: CUDA‑GPU support was added starting from version 0.6.1

Author

Developed by Elieren https://github.com/Elieren .

When using the library, keep an indication of the author.