Ultralytics YOLO LibTorch Inference in C++

A C++ application that runs every Ultralytics YOLO task and model generation as TorchScript with the LibTorch (PyTorch C++ API) and OpenCV. Point it at any exported .torchscript model; the task, class names, and input size are read from the model metadata, and the right post-processing is selected automatically.

✨ Features

All tasks: detect, segment, pose, OBB, classify, and YOLO26 semantic segmentation.
All generations: YOLOv8, YOLO11, and YOLO26. Grid (YOLOv8/11) and end-to-end (YOLO26) outputs are detected automatically.
Zero configuration: task, class names, and imgsz come from the TorchScript config.txt metadata that Ultralytics embeds on export.

📋 Dependencies

Ensure you have the following dependencies installed before proceeding:

Dependency	Version	Resource
OpenCV	>=4.0.0	https://opencv.org/
C++ Standard	>=17	https://isocpp.org/
CMake	>=3.18	https://cmake.org/
Libtorch	>=1.12.1	https://pytorch.org/

You can download the required version of LibTorch from the official PyTorch website. Make sure to select the correct version corresponding to your system and CUDA version (if using GPU).

📦 Exporting a Model

Export any model and task to TorchScript with the Ultralytics export mode:

bash

yolo export model=yolo26n.pt imgsz=640 format=torchscript # detect (also -seg / -pose / -obb / -cls / -sem)

YOLOv8 and YOLO11 grid models work too — the output layout is detected automatically.

See the Export documentation for more options.

🛠️ Build

bash

git clone https://github.com/ultralytics/ultralytics.git
cd ultralytics/examples/cpp/LibTorch
mkdir build && cd build

# Add -DCMAKE_PREFIX_PATH="/path/to/libtorch;/path/to/opencv" if they are not auto-detected.
cmake .. && cmake --build . --config Release

The shared helpers in ../common are header-only and added to the include path automatically.

🚀 Usage

bash

# If LibTorch is not installed system-wide, add its libraries to the loader path:
export LD_LIBRARY_PATH=/path/to/libtorch/lib:$LD_LIBRARY_PATH

# Defaults: --model yolo26n.torchscript --source bus.jpg --conf 0.25 --iou 0.45 --out result.jpg
./yolo_libtorch --model yolo26n.torchscript --source bus.jpg
./yolo_libtorch --model yolo26n-seg.torchscript --source bus.jpg --out seg.jpg
./yolo_libtorch --model yolo26n-pose.torchscript --source bus.jpg --show

Argument	Default	Description
`--model`	`yolo26n.torchscript`	Path to the exported TorchScript model (any task).
`--source`	`bus.jpg`	Input image.
`--conf`	`0.25`	Confidence threshold.
`--iou`	`0.45`	NMS IoU threshold (grid models only).
`--out`	`result.jpg`	Output image path.
`--cuda`	off	Use CUDA if the LibTorch build and a device support it.
`--show`	off	Also open a display window.

The annotated result is always written to --out and the detections are printed to the console. The detected task is shown at startup, e.g. Model: yolo26n.torchscript | task: detect | classes: 80.

🤝 Contributing

Contributions are welcome! If you find any issues or have suggestions for improvements, please feel free to open an issue or submit a pull request on the main Ultralytics repository.