OpenCV 4.0 alpha release out!

1 min read

With more than 3 years from the time of previous version release OpenCV 3.0, the team happily announced the alpha release of the most awaited OpenCV 4.0. The new version is set to encompass exclusive features such as 3D dense reconstruction algorithm, newest improvements and bug fixes to recent maintenance release of OpenCV 3.4.

Key Features:

OpenCV is a C++11 library now
Default features include lambda functions, convenient iteration, and initialization of cv::Mat
Added new chessboard detector
Added exclusive HPX parallel backend and basic FP16 support
Standard std::string and std::shared_ptr replaced hand-crafted cv::String and cv::Ptr
parallel_for can now use the pool of std::threads as backend

Major improvements and bug fixes:

ONNX parser added to existing OpenCV DNN module
Support to various classification networks – AlexNet, Inception v2, Resnet, VGG
Partial support to YOLO v2 object detection network
Faster object detection using Intel Inference Engine, part of Intel OpenVINO
Added stability improvements in the OpenCL backend
Fast QR code detector with support to add QR code decoder soon
SSE4-, AVX2- and NEON-optimized kernels expanded
Legacy C API from OpenCV 1.x partially excluded

This alpha release appears to be a massive version with 85 patches including 28 merge requests. This release is assumed to be quite stable although few changes in OpenCV API and implementation are expected before 4.0 final release. For more information on the detailed list of features and improvements, please read official documentation.