OpenCV 4.10 is out today as the newest version of the Open Computer Vision Library for this widely-used library with machine learning support, object detection, segmentation and recognition, motion video tracking, gesture recognition, and a variety of other features important for today's diverse workloads.

With OpenCV 4.10 there are many improvements to its deep neural network (DNN) module, experimental NumPy 2.0 support, various Android improvements, NVIDIA CUDA 12.4+ platform support, RISC-V and ARM improvements, oneAPI 2024 support, experimental Windows ARM64 support, experimental Apple VisionOS support, and a native Wayland back-end on Linux. Various issues are addressed and improvements made as part of plumbing Wayland support to OpenCV's HighGUI code.

OpenCV 4.10's Deep Neural Network module adds OpenVINO 2024 toolkit support, NVIDIA cuDNN 9+ support, improved modern Yolo detectors, a Vulkan back-end for the NaryEltwiseLayer, parallel version of scatter and scatterND, support for more DNN OpenVINO layers and more ONNX layers, significantly improved DNN memory consumption, and a variety of other optimizations.

Downloads and more information on today's OpenCV 4.10 computer vision release via [1]GitHub .



[1] https://github.com/opencv/opencv/releases/tag/4.10.0