tensorRT_Pro

Why we included this project

tensorRT_Pro wraps TensorRT in a compact C++ and Python API, so a YOLOv5, YOLOX, or YOLOv7 model can go from PyTorch weights to a running inference engine in a few lines of code. It covers the whole workflow, from ONNX export and engine compilation through serialization and CUDA-accelerated preprocessing and postprocessing, and the same wrapper extends to face, pose, and tracking models like RetinaFace, ArcFace, AlphaPose, CenterNet, and DeepSORT. FP32, FP16, and INT8 build paths are supported, and the bundled benchmark logs give you realistic speed numbers to check against your own hardware before committing to a deployment target. The tutorial videos, notebooks, and a commented community fork make it a workable place to learn TensorRT optimization rather than just copy example code.