laugh12321/TensorRT-YOLO

Based on the README, here's a technical summary that goes deeper than the GitHub description: --- Integrates TensorRT custom plugins for accelerated post-processing while leveraging CUDA kernels and CUDA graphs to optimize inference pipelines; supports multiple YOLO variants (v3–v26, YOLO-World, YOLO-Master) across detection, segmentation, classification, pose estimation, and oriented bounding box tasks. Provides single-header C++ API with zero external dependencies and full Python bindings via pybind11, enabling multi-context parallel inference with architecture-specific memory optimizations (e.g., Jetson Zero Copy mode).

1,730 stars. Actively maintained with 6 commits in the last 30 days.