YOLOv8-TensorRT-CPP

Why we included this project

This C++ project shows how to run YOLOv8 on NVIDIA GPUs with TensorRT, handling object detection, semantic segmentation, and body pose estimation. It accepts Ultralytics' pretrained weights directly, so you can point it at a model and get annotated output without writing your own pipeline. The code walks through the full conversion path from PyTorch to ONNX to an optimized TensorRT engine, which is helpful if you want to understand each stage rather than treat it as a black box. Beyond FP32 and FP16, it supports INT8 quantization with calibration data, and includes a benchmark tool that reports preprocessing, inference, and postprocessing times separately. For engineers building low-latency vision applications on Linux with CUDA and OpenCV, this is a practical starting point.