- update to 4.8.1
  * WebP security update for CVE-2023-4863
  * Depthwise convolution 5x5 performance regression fix
- update to 4.8.0, highlights below, for details check
  https://github.com/opencv/opencv/wiki/ChangeLog#version480
  Highlights of this release:
  * DNN module patches:
    + TFLite models support, including int8 quantized models.
    + Enabled DNN module build without Protobuf dependency.
    + Improved layers => supported more models:
      - ONNX: Layer normalization, GELU and QLinearSoftmax.
      - Fixes in CANN backend:
        * support ONNX Split, Slice, Clip (Relu6) and
          Conv with auto_pad.
        * support ONNX Sub, PRelu, ConvTranspose.
      - Reduce Refactor for robustness and potential
        follow-up improvements.
      - Fixes for Segment Anything Model by Meta.
      - Fixes in nary element wise layer about broadcast:
        * Fixes in CPU.
        * and Fixes in CUDA backend.
      - Further increased DNN speed on ARM and X86 by improving
        convolution, covering 1D and 3D cases, supporting
        convolution+element-wise op fusion.
      - Added full FP16 computation branch on ARMv8 platform, 1.5x
        faster than FP32 (FP16 Winograd is still pending).
      - Vulkan backend refactor for better performance and
        robustness. It runs 4X faster than before.
      - Added API blobFromImageParam to build network inputs
        with pre-processings.
      - Modern OpenVINO support.
  * G-API module:
    + Intel® OpenVINO™ inference backend:
      - Streamlined preprocessing in OpenVINO
        Inference Engine (ie) API 1.0 backend. Note: this backend
        will be deprecated after OpenVINO removes the API 1.0
        support in its subsequent releases.
      - Aligned OpenVINO IE API 1.0 backend with the latest
        OpenVINO 2023.0 (as some features were removed there).
      - Introduced a brand new OpenVINO API 2.0 backend.
      - Implemented the required inference operations for
        the OpenVINO API 2.0 backend.
    + Python bindings:
      - Exposed varions normalization options for ONNX RT
        backend in Python bindings.
      - Exposed Fluid kernels and kernel package manipulation
        functions (combine()) in Python.
      - Fixed issues in Stateful Python kernel state
        handling; also fixed various issues in Python tests.
      - Fixed issue with opaque kernel output information
        handling which broke Python custom kernels.
    + Samples:
      - Introduced a new Segmentation demo with desync() to enable
        slow-running networks in the real-time.
      - Updated stats calculation in the G-API-based pipeline
        modelling tool.
    + Other changes and fixes:
      - Fixed tolerance in Fluid resize tests to avoid issues
        on ARM.
      - Fluid backend: extended Merge3 kernel with more
        supported data types.
      - Fixed standalone mode compilation issues.
  * Objdetect module:
    + FaceDetectorYN upgrade for better performance,
      accuracy and facial landmarks support.
    + New QR code detection algorithm based on ArUco code.
    + Bar code detector and decoder moved from Contrib to
      main repository.
    + Introduced common API for all graphical codes like bar
      codes and QR codes.
    + Added flag for legacy pre-4.6.0 ChAruco boards support.
    + Multiple bug fixes and improvements in QR code detection
      and decoding pipelines.
    + Multiple bug fixes and improvements in ArUco based pipelines.
  * Calibration module:
    + USAC framework improvements.
    + Fixed stddev estimation in camera calibration pipelines.
    + Fixed incorrect pixel grid generation in icvGetRectangles
      that improves accuracy of getOptimalNewCameraMatrix,
      stereoRectify and some other calibration functions.
      Charuco board support in patterns generator, interactive
      calibration tool and calibration samples.
  * Image processing module:
    + Various fixes in line segments detector.
    + Fixed even input dimensions for INTER_NEAREST_EXACT
      in resize.
    + Optimise local cost computation in
      IntelligentScissorsMB::buildMap.
    + Keep inliers for linear remap with BORDER_TRANSPARENT
    + Fix distransform to work with large images.
  * Features2d module:
    + SIFT accuracy improvements.
  * Core module:
    + Added REDUCE_SUM2 option to cv::reduce.
    + Introduced cv::hasNonZero function.
    + Update IPP binaries update to version 20230330.
    + Improved RISC-V RVV vector extensions support.
      - Support RVV v0.11 intrinsics available in LLVM 16
        and GCC 13
      - Support build with T-Head RISC-V toolchain
        (RVV 0.7.1 and 1.0)
    + Several OpenCL vendor and version handling improvements.
  * Multimedia:
    + Added AVIF support through libavif.
    + Orbbec Femto Mega cameras support.
    + HEVC/H265 support in VideoWriter with MS Media
      Foundation backend.
    + Fixed FPS computation on some videos for FFmpeg backend.
    + Added support for VideoCapture CAP_PROP_AUTO_WB and
      CV_CAP_PROP_WHITE_BALANCE_BLUE_U for DShow backend.
    + Fixes OBS Virtual Camera capture.
    + CV_32S encoding support with tiff.
  * Python Bindings:
    + Python typing stubs.
    + Fix reference counting errors in registerNewType.
    + Fixed ChAruco and diamond boards detector bindings.
    + Added bindings to allow GpuMat and Stream objects to
      be initialized from memory initialized in other libraries
    + np.float16 support.
    + Python bindings for RotatedRect, CV_MAKETYPE, CV_8UC(n).
  * JavaScript bindings:
    + Added possibility for disabling inlining wasm in opencv.js
    + Extended JS bindings for Aruco, Charuco, QR codes and
      bar codes.
  * Other:
    + Several critical issue fixes in wechat_qrcode module
      (opencv_contrib)

• Files Changed
• Browse Source