OpenCV 5.1 adds AVX-512, RISC-V vector kernels, and FP8 support

The OpenCV project has released version 5.1, providing a series of incremental improvements and bug fixes to its widely used computer vision library. This update focuses heavily on hardware acceleration for modern AI workloads and diverse processor architectures.

Key changes include enhanced SIMD optimizations for x86_64 systems and new vector kernels for RISC-V. The release also streamlines the library by removing the legacy DNN engine in favor of the redesigned engine introduced in version 5.0.

Technical improvements in OpenCV 5.1

• SIMD and Vectorization: The release introduces new AVX-512 and AVX-VNNI optimizations for x86 architectures, as well as new RISC-V RVV vector kernels.
• AI and Data Types: Support for FP8 has been expanded across various components, and the DNN module received improvements for NVIDIA CUDA and cuDNN.
• Engine Consolidation: The legacy DNN engine has been officially removed to consolidate development on the new engine debuted in OpenCV 5.0.
• Visualization and Bindings: A new OpenGL-based 3D visualizer is available for meshes, point clouds, and camera trajectories, alongside improved bindings for Python, JavaScript, and Java.

Why it matters

By optimizing for AVX-512 and RISC-V RVV, OpenCV 5.1 ensures high-performance computer vision capabilities on both high-end workstations and emerging open hardware platforms. The expanded FP8 and CUDA support further strengthens its utility for developers building AI-driven applications on Linux-based edge and cloud systems.