xlite-dev/lite.ai.toolkit

Supports inference across multiple backend runtimes (ONNX Runtime, MNN, TensorRT, NCNN, TNN) through a unified C++ API, enabling flexible deployment from CPU to GPU without code changes. Features 300+ pre-implemented model classes with standardized syntax (`lite::cv::Type::Class`) and minimal dependencies—only OpenCV and ONNXRuntime required by default. Includes pre-trained weights across 500+ variants accessible through dedicated model hubs, facilitating rapid prototyping for computer vision tasks including detection, segmentation, matting, and face operations.

4,387 stars. Actively maintained with 1 commit in the last 30 days.