OpenCV 5 launches with rewritten DNN engine

By shedding its decades-old legacy baggage and focusing on modern deep learning infrastructures, OpenCV 5 shifts from a classic image processing utility into a highly optimized runtime for modern edge AI and generative vision models.

• –The rewritten DNN engine and 80%+ ONNX support make OpenCV a much more formidable, lightweight alternative to ONNX Runtime and TensorRT for edge deployments.
• –Requiring C++17 and completely removing the C API is a long-overdue housecleaning that will significantly improve developer experience and codebase maintainability.
• –Moving older components like G-API and the ML module to opencv_contrib ensures the core library remains nimble and lightweight.
• –The addition of native LLM/VLM support prepares the library for the multi-modal future, although it remains to be seen how developers will balance OpenCV with specialized LLM inference frameworks.