Real-Time Object Detection / Segmentation in C++ with YOLO and Onnx Runtime

An example of using Yolo with the Onnx runtime in C++ that goes along with my blog post on Medium.

Requirements:

• ONNX Runtime
• OpenCV
• Cmake
• C++ compiler
• CUDA Toolkit / cuDNN (GPU Only)

See the object detection article on Medium for more details.

Project Structure:

• c++ - Contains the C++ example code for object detection and segmentation using ONNX runtime.
• python - Contains the python scripts for converting pytorch weights to ONNX format and object segmentation using ONNX runtime.
• images - Contains sample images used in the blog post.
• labels - Contains the class names for the YOLO model trained on COCO dataset.
• weights - Contains the weights for the YOLO object detection / segmentation model trained on COCO dataset from Ultralytics.

Speed Test

Here is the breakdown of the results of the speed test:

GPU: GeForce RTX 4070

Model: YOLOv11n

C++

Filename: video_test.cpp

Task	Average Latency (ms)
Pre-processing	6
Inference	6
Bounding Box Post-processing	1
Mask Post-processing	3
Drawing Boxes / Masks	1
Total loop time	20

Frames per second
50

Python (Not finished yet)

Filename: video_test.cpp

Task	Average Latency (ms)
Pre-processing
Inference
Bounding Box Post-processing
Mask Post-processing
Drawing Boxes / Masks
Total loop time

Frames per second