v0.1.33

Changelog

v0.1.33 (2023-08-01)

Browse the Repository | Released Assets

What's Changed

• Updated to OpenCV 4.8.0. Some APIs may have changed, please see OpenCV's release note for more information.
• Using manylinux2014 to build precompiled binaries for x86_64-linux-gnu (w/ and w/o contrib, except for the cuda ones). This lowers the required glibc version to 2.17.
• docs: fix constants by @mnishiguchi in #190
• docs(warpPerspective): enhance with Kino and some text by @mnishiguchi in #191
• Correct the release dates in CHANGELOG.md by @edwardzhou in #194
• override env EVISION_PREFER_PRECOMPILED when building from source by @cocoa-xu in #197
• Using asdf by @cocoa-xu in #199
• Fix model zoo url by @RyoWakabayashi in #204
• Fix face detection bbox by @RyoWakabayashi in #207
• Update model zoo URL by @RyoWakabayashi in #205
• [model-zoo/sface] fix face recognition by @cocoa-xu in #208
• nif 2.17 by @cocoa-xu in #202
• update to OpenCV 4.8.0 by @cocoa-xu in #213

New Contributors

• @mnishiguchi made their first contribution in #190
• @RyoWakabayashi made their first contribution in #204

Full Changelog: v0.1.31...v0.1.33

v0.1.31

Changelog

v0.1.31 (2022-04-17)

Browse the Repository | Released Assets

Changed

• [Evision.DNN] Added support for passing Nx.Tensor or Evision.Mat as the input argument of bboxes in

  • Evision.DNN.nmsBoxes/{4,5}

    iex> Evision.DNN.nmsBoxes([{0,1,2,3}], [1], 0.4, 0.3)
    [0]
    iex> Evision.DNN.nmsBoxes(Nx.tensor([[0,1,2,3]]), [1], 0.4, 0.3)
    [0]
    iex> Evision.DNN.nmsBoxes(Evision.Mat.literal([[0,1,2,3]], :f64), [1], 0.4, 0.3)
    [0]

  • Evision.DNN.nmsBoxesBatched/{5,6}

    iex> Evision.DNN.nmsBoxesBatched([{0,1,2,3}], [1], [1], 0.4, 0.3)
    [0]
    iex> Evision.DNN.nmsBoxesBatched(Nx.tensor([[0,1,2,3]]), [1], [1], 0.4, 0.3)
    [0]
    iex> Evision.DNN.nmsBoxesBatched(Evision.Mat.literal([[0,1,2,3]], :f64), [1], [1], 0.4, 0.3)
    [0]

  • Evision.DNN.softNMSBoxes/{4,5}

    iex> Evision.DNN.softNMSBoxes([{0,1,2,3}], [1], 0.4, 0.3)
    {[1.0], [0]}
    iex> Evision.DNN.softNMSBoxes(Nx.tensor([[0,1,2,3]]), [1], 0.4, 0.3)
    {[1.0], [0]}
    iex> Evision.DNN.softNMSBoxes(Evision.Mat.literal([[0,1,2,3]], :s32), [1], 0.4, 0.3)
    {[1.0], [0]}

v0.1.30

v0.1.30

v0.1.29

Changelog

v0.1.29 (2022-03-13)

Browse the Repository | Released Assets

Fixed

• [py_src] Fixed incorrect typespec for Saclar. Thanks @tusqasi and @Nicd.
• [smartcell] Fixed PPResnet based models.
• [smartcell] Fixed invalid charset URLs as they were removed in the upstream repo. CRNN models URLs to commit 12817b80.

New Contributors

• @Nicd made their first contribution in #182

Full Changelog: v0.1.28...v0.1.29

v0.1.28

v0.1.28 (2022-01-25)

Browse the Repository | Released Assets

Fixed

• [py_src] ArgInfo.has_default is now set to true if a.defval is f"{a.tp}()". Fixed #174.
• [ci-win-precompile-core] Removed the line that deletes the priv/x64 directory. It should be removed in the last version because we will add priv/x64 to the DLL search path instead of copying all opencv dll files to the priv directory.

v0.1.27

v0.1.27 (2022-01-23)

Browse the Repository | Released Assets

Breaking Changes

• [Evision.Constant] Constant values are all relocated to the Evision.Constant module. To use them, either do

  Evision.Constant.cv_IMREAD_ANY()

  or

  import Evision.Constant
  cv_IMREAD_ANY()

• [Evision] Evision.__enabled_modules__/0 => Evision.enabled_modules/0. Result will now be computed using HAVE_OPENCV_{MODULE_NAME} macros.

Fixed

• [Evision.Mat] fixed Evision.Mat.update_roi/3.
• [py_src] fix incorrect typespecs.
• [py_src] VideoCaptureAPIs should be a single number instead of a list of number.

Changed

• [c_src] check if we can use existing atom from enif_make_existing_atom before calling to enif_make_atom.

Added

• OpenCV contrib modules.

  • For users who prefer using precompiled binaries, almost all modules in opencv_contrib is included in precompiled archives (except for CUDA related ones, please see the next bullet point), and this will be the new default for precompiled users.

    However, we do provide precompiled binaries that only include core modules. Please see the next paragraph.

  • For all users, to use core modules only, please set env var EVISION_ENABLE_CONTRIB=false.

• CUDA 11 + cudnn 8 support with precompiled binaries.

  Note that CUDA 11 and cudnn 8 runtime libraries/dll files are not included in the precompiled archive.

  Please following the installation guide on NVIDIA's website.

  • For precompiled binaries users, please set EVISION_ENABLE_CUDA to true. Besides that, there are 3 CUDA versions to choose:

    • EVISION_CUDA_VERSION=111
    • EVISION_CUDA_VERSION=114
    • EVISION_CUDA_VERSION=118

    Please choose the one that matches your local CUDA versions, and set this EVISION_CUDA_VERSION env var correspondingly.

  • For users who prefer compiling from source, you'll only need to set EVISION_ENABLE_CUDA to true, and OpenCV will detect and use (if possible) your local CUDA/cudnn runtime.

  Lastly, if EVISION_ENABLE_CUDA is true while EVISION_ENABLE_CONTRIB is false, CUDA related modules will not be compiled/downloaded.

v0.1.26-rc3

Merge pull request #169 from cocoa-xu/cx-cuda

CUDA support

v0.1.26-rc2

[script] fixed a typo

v0.1.26-rc1

bump all

v0.1.26-rc0

set correct elixir module name in `evision_${name}_Instance`