laugh12321/TensorRT-YOLO

What is it?

Key features

• 01Supports detect, segment, classify, pose and OBB tasks for YOLO-family models.
• 02Provides both C++ and Python inference APIs with the same pipeline.
• 03Uses TensorRT plugins for post-processing and CUDA kernels plus CUDA Graphs for speed.
• 04Offers a CLI that detects model structure automatically.
• 05The C++ module is a single header with no third-party dependencies.

Use cases

• Deploying real-time YOLO object detection on NVIDIA GPUs or Jetson devices.
• Running instance segmentation or pose estimation services with TensorRT engines.
• Serving multiple YOLO models in C++ applications with multi-context inference.
• Building video analytics pipelines on NVIDIA hardware.

Quick start

1. 1

   Clone the TensorRT-YOLO repository

   git clone https://github.com/laugh12321/TensorRT-YOLO
2. 2

   Install pybind11 for Python bindings

   pip install "pybind11[global]"
3. 3

   Configure the build with CMake, pointing TRT_PATH to your TensorRT directory

   cmake -S . -B build -D TRT_PATH=/your/tensorrt/dir -D BUILD_PYTHON=ON -D CMAKE_INSTALL_PREFIX=/your/tensorrt-yolo/install/dir
4. 4

   Build and install the library

   cmake --build build -j$(nproc) --config Release --target install

Things to know

• !Requires an NVIDIA GPU with CUDA and TensorRT installed (TensorRT 8.6.1 or newer recommended).
• !Licensed under GPL-3.0, a copyleft license that requires derivative works to be open-sourced under the same terms.
• !The community edition's inference interface is frozen and will not receive new runtime architecture; the README describes a closed-source professional edition for advanced features.

In collections

Topics

FAQ

What is TensorRT-YOLO?

A C++/Python toolkit for deploying YOLO family models on NVIDIA GPUs with TensorRT, covering detect, segment, classify, pose and OBB tasks. TensorRT-YOLO is a deployment toolkit for running YOLO-family models on NVIDIA hardware through TensorRT. It provides C++ and Python inference interfaces for object detection, instance segmentation, image classification, pose estimation, and rotated object detection, and includes a video analysis example.

Is TensorRT-YOLO free to use? What license does it have?

TensorRT-YOLO is open source under the GPL-3.0 license. Check the license terms before using it in commercial products.

Is TensorRT-YOLO still maintained?

Updated within 6 months. The last commit to TensorRT-YOLO was on Sep 10, 2026.

What can I use TensorRT-YOLO for?

Deploying real-time YOLO object detection on NVIDIA GPUs or Jetson devices. Running instance segmentation or pose estimation services with TensorRT engines. Serving multiple YOLO models in C++ applications with multi-context inference. Building video analytics pipelines on NVIDIA hardware.

How do I install TensorRT-YOLO?

Clone the TensorRT-YOLO repository: git clone https://github.com/laugh12321/TensorRT-YOLO Install pybind11 for Python bindings: pip install "pybind11[global]" Configure the build with CMake, pointing TRT_PATH to your TensorRT directory: cmake -S . -B build -D TRT_PATH=/your/tensorrt/dir -D BUILD_PYTHON=ON -D CMAKE_INSTALL_PREFIX=/your/tensorrt-yolo/install/dir Build and install the library: cmake --build build -j$(nproc) --config Release --target install

Where can I get TensorRT-YOLO?

The source code and installation instructions are on GitHub: https://github.com/laugh12321/TensorRT-YOLO