Difference between revisions of "Version History"

From Emgu CV: OpenCV in .NET (C#, VB, C++ and more)
Jump to: navigation, search
(Change Log)
 
(36 intermediate revisions by the same user not shown)
Line 1: Line 1:
 +
==Emgu.CV-4.5.1==
 +
===Online Documentation===
 +
[{{SERVER}}/wiki/files/4.5.1/document/index.html Browse]
 +
 +
===Change Log===
 +
* 4.5.1 is based on the [https://github.com/opencv/opencv/wiki/ChangeLog#version451 Open CV 4.5.1 release].
 +
* Changes in nuget packages:
 +
**  '''Emgu.CV.runtime.windows.cuda package for this release is not available from nuget.org'''. If you need to use the ''Emgu.CV.runtime.windows.cuda'' package, you can download two large nuget packages ( [http://www.emgu.com/wiki/files/4.5.1/release/Emgu.CV.runtime.windows.cuda.4.5.1.4349.nupkg Emgu.CV.runtime.windows.cuda.4.5.1.4349.nupkg]; [http://www.emgu.com/wiki/files/4.5.1/release/Emgu.runtime.windows.cuda.dnn.cnn.8.0.4.nupkg Emgu.runtime.windows.cuda.dnn.cnn.8.0.4.nupkg] ) and hosted it on your local nuget repository. The latest CUDA DNN package has a file "cudnn_cnn_infer64_8.dll" in v8.0.4 that is 688MB in size, after the nuget package compression the nupkg file is still 287MB. Nuget.org has a hard limit in package size of 250MB, this prevent the package being uploaded to nuget.org. Unless NVidia break up "cudnn_cnn_infer64_8.dll" into multiple smaller files, or nuget.org increase the file limit, we will not be able to upload the latest cuda nuget packages into nuget.org. The existing 4.4 release use older version of CUDNN and has nuget package reaching 248MB, we have moved onto newer version of CUDA and CUDNN to support the RTX 30xx series of graphic card and is no longer using the previous version of CUDNN. '''Commercial license holder will not be affected''', we have the Emgu.CV.runtime.windows.cuda package uploaded to our commercial nuget repository which have larger nupkg size limit (500MB).
 +
** Renamed ''Emgu.CV.runtime.windows.cuda'' package to ''Emgu.runtime.windows.cuda''. This package is now a common dependency with ''Emgu.TF.runtime.windows.cuda'' nuget package.
 +
** Moved the msvc runtime out from ''Emgu.CV.runtime.windows'' package to two separate packages: ''Emgu.runtime.windows.msvc.rt.x86'' and ''Emgu.runtime.windows.msvc.rt.x64''.
 +
** Renamed the ''Emgu.CV.runtime.ubuntu'' package to ''Emgu.CV.runtime.ubuntu.20.04-x64'' to better reflect the native binary the package contains.
 +
** Renamed the ''Emgu.CV.runtime.raspbian'' package to ''Emgu.CV.runtime.linux-arm''. Raspberry pi foundation has renamed ''raspbian'' to ''raspberry os'' and we are moving away from the name ''raspbian'' for this reason.
 +
* The Mac OS native fat binary now contains native ARM64 slice for the apple silicon (as well as x86-64). As soon as .net is available for apple silicon, you will be able to run emgu cv in native arm64 mode for the new Mac.
 +
* iOS native binary is build with XCode 12. Removed x86 simulator from the native iOS binary. x86 simulator is no longer supported in XCode 12.
 +
* CUDA release is now build with CUDA 11.1 and CUDNN 8, the highest CUDA bin option has been increased to 8.6 for RTX 3xxx series. The minimum CUDA bin option has been lowered to 5.2 for older device.
 +
* Windows commercial release now build with OpenVino 2021
 +
* Added MCC module
 +
* Added HDF module
 +
* Enabled Alphamat module for Mac OS and Ubnutu
 +
* Re-enable CudaCascadeClassifier properties.
 +
* Increase IPHONEOS_DEPLOYMENT_TARGET from 8.0 to 9.0 (for XCode 12).
 +
* Added function in BitmapExtension to convert Bitmap to Mat.
 +
* Added VectorOfSize class
 +
* Added TextDetectorCNN class in Emgu.CV.Text module
 +
* Added RetrType.Floodfill for FindContours function.
 +
* Added HDDL target for openvino.
 +
* Renamed CheckLibraryLoaded() to Init()
 +
 +
 +
====Enhancements====
 +
* Added FreeType demo to Xamarin Forms demo App.
 +
* Use DisparityWLSFilter for Simple3DReconstruct demo.
 +
* Updated demo projects that depends on .Net Core 3.1 to .Net 5
 +
 +
====Bug Fixes====
 +
* Fixed a bug where lowerBound is not passed to EMD.
 +
* Fixed [https://github.com/emgucv/emgucv/issues/370 issue 370]: method CalibrateHandEye using wrong input array.
 +
* Fixed [https://github.com/emgucv/emgucv/issues/393 issue 393]: a bug in IDisparityFilter.Filter function.
 +
* Fixed Emgu.CV.runtime.ubuntu and Emgu.CV.runtime.raspbian nuget package, such that the native binary are correctly deployed.
 +
* Fixed [https://github.com/emgucv/emgucv/issues/389 issue 389]: UWP compilation with .net native toolchain.
 +
* For windows, when loading native binary, Emgu CV no longer change Enviroment.CurrentDirectory
 +
 +
 +
==== Dependency ====
 +
* Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack. e.g. for windows server 2012 R2, following [https://github.com/emgucv/emgucv/issues/206 this instruction]
 +
 +
==Emgu.CV-4.4.0==
 +
===Online Documentation===
 +
[{{SERVER}}/wiki/files/4.4.0/document/index.html Browse]
 +
 +
===Change Log===
 +
* 4.4.0 is based on the Open CV 4.4.0 release.
 +
* The standard license terms has been change to dual license from GPL. See it here: https://github.com/emgucv/emgucv/blob/master/LICENSE The change is to allow commercial license owners to use our regular open source build and the nuget packages from nuget.org.
 +
* Change in nuget packages:
 +
** Both commercial and open source nuget package now use the Emgu.CV nuget package from nuget.org. The commercial release provide runtime nuget packages that are built with OpenVino and have IPP and CPU dispatch enabled.
 +
** No longer append "-CR" to commercial nuget package. Add "dldt" to nuget package compiled with openvino. This is to make clear what was in the commercial release and matches the Open CV package naming scheme (such as opencv-4.4.0-dldt-2020.4-vc16-avx2.zip). For example, the Emgu.CV.windows.runtime-CR package in v4.3 has been renamed to Emgu.CV.windows.runtime.dldt in v4.4. The Emgu.CV.windows.runtime.cuda-CR package in v4.3 has been renamed to Emgu.CV.windows.runtime.cuda.dldt in v4.4.
 +
** Added Emgu.CV.Bitmap nuget package that contains the conversion code from Emgu CV images to Bitmap. This package targets .NetStandard platform. If you are running on Linux / Mac OS platform, you will need to make sure native GDI+ is installed in your system, or it will failed on dependency when Bitmap is used.
 +
** Added Emgu.CV.UI nuget package. This package contains Emgu.CV.Platform.NetFramework.dll and Emgu.CV.Platform.NetCore.dll originally contained in Emgu.CV nuget pacakge in v4.3. These two dlls are only compatible with Windows platform.
 +
** Emgu.CV nuget package now only targets .NetStandard platform.
 +
** Split up Emgu.CV.runtime.cuda.dnn nuget packages into 3: Emgu.CV.runtime.cuda.dnn.cnn, Emgu.CV.runtime.cuda.dnn.ops & Emgu.CV.runtime.cuda.dnn. No changes is needed on the user side, the dependency resolution will be handled automatically by nuget package manager.
 +
* The 64-bit commercial binary for Windows is compiled with Intel Inference Engine backend v2020.4 (instead of v2020.3 in the v4.3 release)
 +
* Enable IPP and CPU dispatch for windows commercial build.
 +
* For Android runtime, in v4.3 we switched to from Open CV's custom toolchain to the android toolchain from NDK. There is a bug in Android NDK that enabled debug symbols even in release built. As a result, the Android binary is ~6 times larger. We have updated our build flags to force to exclude the debug symbols in release built. In v4.4, the android binary size are back to the v4.2 level.
 +
* 4.4.0 release is build with Tesseract 4.1.1
 +
* GPU release is build with CUDA 11.0 instead of CUDA 10
 +
* Added Intensity Transform module
 +
* Added Rapid module
 +
* Added eigen3 submodule to build alphamat on windows and android.
 +
* Enable saliency module for Android.
 +
* Re-enable CudaCascadeClassifier class.
 +
* Enable building DynamicImage.Asp.NetCore project on Linux, Mac and Raspbian.
 +
* Added HelloWorld.UWP project for UWP build.
 +
* Added CvInvoke.ScaleAdd function.
 +
* Added build script for Ubuntu 20.04
 +
* VideoCapture.Read function now return bool instead of void.
 +
* Updated XamarinForms.WPF to base on .NetCore instead of .Net Framework 4.7
 +
* Updated WPF demo project to base on .NetCore instead of .Net Framework 4.6.1
 +
* Change DnnInvoke.GetAvailableBackends() to property DnnInvoke.AvailableBackends
 +
* Moved SIFT from XFeatures2D to Features2D namespace
 +
* VideoCapture.Run function use Emgu.Util.ExceptionHandler instead of System.ServiceModel.Dispatcher.ExceptionHandler.
 +
* The License Plate detection demo now use DNN with OpenVino model zoo for improved detection accuracy.
 +
 +
 +
====Enhancements====
 +
* Implemented CI build on github.
 +
 +
====Bug Fixes====
 +
* This version can run on workstation without AVX2 instruction set. Fixed the known issue for Open CV 4.3 where the dldt build is not able to run on workstations without AVX2.
 +
 +
==== Dependency ====
 +
* Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack. e.g. for windows server 2012 R2, following [https://github.com/emgucv/emgucv/issues/206 this instruction]
 +
 +
==Emgu.CV-4.3.0==
 +
===Online Documentation===
 +
[{{SERVER}}/wiki/files/4.3.0/document/index.html Browse]
 +
 +
===Change Log===
 +
* 4.3.0 is based on the Open CV 4.3.0 release.
 +
* For commercial release, we no longer offer binary compiled with Intel C++ compiler. In many case, Visual C++ is producing binaries as efficient as those produced with Intel C++ compiler. In certain situations (e.g. DNN modules running Facial Landmark demo), the binary produce by Visual C++ even run faster than those produced by C++ compiler. We decided to build our 64-bit commercial release without Intel C++, and instead, including Intel Inference Engine for better performance on the DNN module.
 +
* The 64-bit commercial binary for Windows is compiled with Intel Inference Engine backend v2020.2. This version required a machine with AVX2 instruction set, or else it will crash when DNN module is loaded. See [[Version_History#Known_Issues | Known Issues]]. Official Open CV 4.3 release has the exact same bug from Open Vino.
 +
* 4.3.0 release is build with Tesseract 4.1.1
 +
* Mac OS binary is now signed using our developer certificate.
 +
* Android release is now built with NDK r21. Android demo now has a minSdkVersion="24" & targetSdkVersion="29"
 +
* Added runtime shared project for Raspbian and Ubuntu
 +
* Ubuntu and Raspbian build script now require dotnet core 3.1 instead of mono
 +
* Enable building raspbian and Ubuntu nuget pacakge.
 +
* Removed ZedGraph dependency from Emgu.CV.UI. HistogramBox now use Open CV's plot module to draw histogram instead of using ZedGraph.
 +
* Removed CvBlob. It has not been updated for 7 years. Please use opencv's tracking module instead.
 +
* SIFT detector is enabled by default due patents expiration (without requirement of NONFREE build option)
 +
* Added Emgu.CV.UI.NetCore project which provide a .NetCore compatible implementation of Emgu.CV.UI
 +
* Added Emgu.CV.Cuda.NvidiaOpticalFlow_1_0. Only support NVidia Graphic cards with Turing architecture (e.g. RTX 20xx)or above.
 +
* Added SetWindowProperty, GetWindowProperty & SetWindowTitle to CvInvoke. Renamed NamedWindowType to WindowFlags.
 +
* Added BundleAdjuster, ExposureCompensator, SeamFinder, Estimator and FeaturesMatcher for stitching.
 +
* Added CylindricalWarper, CylindricalWarperGpu, PlaneWarperGpu & SphericalWarperGpu for stitching.
 +
* Added RotationWarper for stitching.
 +
* Added CvInboke.EstimateChessboardSharpness
 +
* Added warpImage, estimateTransformation & applyTransformation function for IShapeTransformer
 +
* Added h/vconcate overload that accept multiple Mats.
 +
* Added LBPHFaceRecognizer.Histograms property.
 +
* Rename HoughTypes to HoughModes. Added CvEnum.HoughMode.GradientAlt.
 +
* Added default value for FeatherBlender's "sharpness" parameter.
 +
* Added "Values" property for VectorOfObject debugger visualizer, when it is not a VectorOfVector.
 +
* Turned HelloWorld demo from .NetFramework console app to .NetCore console app.
 +
* Added BuildInfo.NetCore.Console demo project.
 +
* Use Xamarin.Plugin.FilePicker to pick file on Xamarin Forms Windows WPF demo.
 +
* Use MaskRcnn with DNN module for TrafficSignRecognition demo.
 +
* Apply Non-maximum suppression on MaskRcnn demo.
 +
* Updated VideoSurveillance demo to use Tracking module instead of CvBlob.
 +
* Replaced WebDynamicImage demo with DynamicImage.Asp.NetCore demo.
 +
* Moved ShapeDetection demo to Xamarin Forms.
 +
* Emgu.Util.Platform can now detect iOS platform.
 +
* Removed lots of __IOS__ conditional compilation flags in Emgu CV source code.
 +
* Use .net core 3.1 for unit test.
 +
 +
====Enhancements====
 +
* Added Yolo demo project in Xamarin Forms.
 +
* Demo projects now use async function calls to download model files from Internet.
 +
* Xamarin forms Android demo (Face Detection, Face Landmark Detection, Mask Rcnn Detection, Stop Sign detection, Yolo) can now use Camera in real time.
 +
 +
====Bug Fixes====
 +
* Fixed a bug in WText Dispose function.
 +
* Fixed a bug in cveSuperresFrameSourceRelease PInvoke.
 +
* Fixed cveSuperResolutionRelease header definition.
 +
* Fixed a bug in QuasiDenseStereo.Process implementation.
 +
* Fixed the spelling on cvePlaneWarperCreate.
 +
* Fixed a bug in loading image file into Image<Bgr, double>.
 +
 +
==== Dependency ====
 +
* Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack. e.g. for windows server 2012 R2, following [https://github.com/emgucv/emgucv/issues/206 this instruction]
 +
 +
==== Known Issues====
 +
* Debugger Visualizer does not work in this release. Getting debugger visualizer working for .NetStandard profile is a much challenging task than we originally thought. We will try to address that in the next release.
 +
* For commercial release, it is build with Intel Inference Engine backend for Windows. Due to a [https://github.com/openvinotoolkit/openvino/issues/387 bug in OpenVINO], the application will crashed on Machines without AVX2 instructions sets. This bug should be fixed in OpenVINO 2020.4 according to the ticket.
 +
 
==Emgu.CV-4.2.0==
 
==Emgu.CV-4.2.0==
 
===Online Documentation===
 
===Online Documentation===
Line 5: Line 159:
 
===Change Log===
 
===Change Log===
 
* 4.2.0 is based on the Open CV 4.2.0 release
 
* 4.2.0 is based on the Open CV 4.2.0 release
 +
* The Emgu.CV dll is now targeting .NetStandard 2.0 instead of .Net 3.5
 +
* The Emgu.CV.UI dll is now targeting .Net 4.6.1 instead of .Net 3.5
 +
* '''The nuget package structures has been changed.''' Emgu.CV nuget package < 4.2.0 contains both the .Net 3.5 dll and the native dll for Windows. Starting from 4.2.0, Emgu.CV nuget package will contain only the .NetStandard 2.0 profile dll. The native dll comes in separate packages for different platforms. e.g. The Emgu.CV.runtime.windows package contains the native dll for Windows as well as the Emgu.CV.UI dll for Windows when targeting .netframework 4.6.1+. The Emgu.CV-CUDA nuget package has been replaced with Emgu.CV.runtime.windows.cuda nuget package.
 +
* Open CV CUDA DNN module required Compute 5.3 and higher. Our cuda build not longer build with cuda compute bin option < 5.3.
 +
* Conversion between Mat, UMat, GpuMat and Image<,> and Bitmap objects requires code changes. The constructors of the Mat, UMat, GpuMat and Image<,> that accepts Bitmap has been removed. It has been added to the Emgu.CV.UI.dll file as an extension method. The dll is part of the Emgu.CV.runtime.windows (or the similar Emgu.CV.runtime.windows.cuda) nuget pacakge. Prior to v4.2.0 this code can be use to create a Mat from Bitmap <pre>Mat m = new Mat(bitmap);</pre>. In the 4.2.0 release it has to be written like this <pre>Mat m = bitmap.ToMat()</Pre> For <pre>Image<,>.Bitmap</pre> property in 4.1.x and older release, its equivalent in v4.2.0 is <pre>Image<,>.AsBitmap()</pre>
  
 
====Enhancements====
 
====Enhancements====
 +
* Added DnnSuperres module.
 +
* Added the turbo color map.
 +
* Added Vulkan and Cuda backend enumeration for DNN.
 +
* Added FSR Inpainting type.
 +
* Added Features2DToolbox.DrawMatches overload.
 +
* Added CvInvoke.DrawMarker
  
 
====Bug Fixes====
 
====Bug Fixes====
 +
* Fixed the ambiguous reference compilation error between 'Emgu.CV.Structure.Range' and 'System.Range' to make it compatible with the latest version of Visual Studio
  
 
==== Dependency ====
 
==== Dependency ====

Latest revision as of 19:10, 20 January 2021

Emgu.CV-4.5.1

Online Documentation

Browse

Change Log

  • 4.5.1 is based on the Open CV 4.5.1 release.
  • Changes in nuget packages:
    • Emgu.CV.runtime.windows.cuda package for this release is not available from nuget.org. If you need to use the Emgu.CV.runtime.windows.cuda package, you can download two large nuget packages ( Emgu.CV.runtime.windows.cuda.4.5.1.4349.nupkg; Emgu.runtime.windows.cuda.dnn.cnn.8.0.4.nupkg ) and hosted it on your local nuget repository. The latest CUDA DNN package has a file "cudnn_cnn_infer64_8.dll" in v8.0.4 that is 688MB in size, after the nuget package compression the nupkg file is still 287MB. Nuget.org has a hard limit in package size of 250MB, this prevent the package being uploaded to nuget.org. Unless NVidia break up "cudnn_cnn_infer64_8.dll" into multiple smaller files, or nuget.org increase the file limit, we will not be able to upload the latest cuda nuget packages into nuget.org. The existing 4.4 release use older version of CUDNN and has nuget package reaching 248MB, we have moved onto newer version of CUDA and CUDNN to support the RTX 30xx series of graphic card and is no longer using the previous version of CUDNN. Commercial license holder will not be affected, we have the Emgu.CV.runtime.windows.cuda package uploaded to our commercial nuget repository which have larger nupkg size limit (500MB).
    • Renamed Emgu.CV.runtime.windows.cuda package to Emgu.runtime.windows.cuda. This package is now a common dependency with Emgu.TF.runtime.windows.cuda nuget package.
    • Moved the msvc runtime out from Emgu.CV.runtime.windows package to two separate packages: Emgu.runtime.windows.msvc.rt.x86 and Emgu.runtime.windows.msvc.rt.x64.
    • Renamed the Emgu.CV.runtime.ubuntu package to Emgu.CV.runtime.ubuntu.20.04-x64 to better reflect the native binary the package contains.
    • Renamed the Emgu.CV.runtime.raspbian package to Emgu.CV.runtime.linux-arm. Raspberry pi foundation has renamed raspbian to raspberry os and we are moving away from the name raspbian for this reason.
  • The Mac OS native fat binary now contains native ARM64 slice for the apple silicon (as well as x86-64). As soon as .net is available for apple silicon, you will be able to run emgu cv in native arm64 mode for the new Mac.
  • iOS native binary is build with XCode 12. Removed x86 simulator from the native iOS binary. x86 simulator is no longer supported in XCode 12.
  • CUDA release is now build with CUDA 11.1 and CUDNN 8, the highest CUDA bin option has been increased to 8.6 for RTX 3xxx series. The minimum CUDA bin option has been lowered to 5.2 for older device.
  • Windows commercial release now build with OpenVino 2021
  • Added MCC module
  • Added HDF module
  • Enabled Alphamat module for Mac OS and Ubnutu
  • Re-enable CudaCascadeClassifier properties.
  • Increase IPHONEOS_DEPLOYMENT_TARGET from 8.0 to 9.0 (for XCode 12).
  • Added function in BitmapExtension to convert Bitmap to Mat.
  • Added VectorOfSize class
  • Added TextDetectorCNN class in Emgu.CV.Text module
  • Added RetrType.Floodfill for FindContours function.
  • Added HDDL target for openvino.
  • Renamed CheckLibraryLoaded() to Init()


Enhancements

  • Added FreeType demo to Xamarin Forms demo App.
  • Use DisparityWLSFilter for Simple3DReconstruct demo.
  • Updated demo projects that depends on .Net Core 3.1 to .Net 5

Bug Fixes

  • Fixed a bug where lowerBound is not passed to EMD.
  • Fixed issue 370: method CalibrateHandEye using wrong input array.
  • Fixed issue 393: a bug in IDisparityFilter.Filter function.
  • Fixed Emgu.CV.runtime.ubuntu and Emgu.CV.runtime.raspbian nuget package, such that the native binary are correctly deployed.
  • Fixed issue 389: UWP compilation with .net native toolchain.
  • For windows, when loading native binary, Emgu CV no longer change Enviroment.CurrentDirectory


Dependency

  • Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack. e.g. for windows server 2012 R2, following this instruction

Emgu.CV-4.4.0

Online Documentation

Browse

Change Log

  • 4.4.0 is based on the Open CV 4.4.0 release.
  • The standard license terms has been change to dual license from GPL. See it here: https://github.com/emgucv/emgucv/blob/master/LICENSE The change is to allow commercial license owners to use our regular open source build and the nuget packages from nuget.org.
  • Change in nuget packages:
    • Both commercial and open source nuget package now use the Emgu.CV nuget package from nuget.org. The commercial release provide runtime nuget packages that are built with OpenVino and have IPP and CPU dispatch enabled.
    • No longer append "-CR" to commercial nuget package. Add "dldt" to nuget package compiled with openvino. This is to make clear what was in the commercial release and matches the Open CV package naming scheme (such as opencv-4.4.0-dldt-2020.4-vc16-avx2.zip). For example, the Emgu.CV.windows.runtime-CR package in v4.3 has been renamed to Emgu.CV.windows.runtime.dldt in v4.4. The Emgu.CV.windows.runtime.cuda-CR package in v4.3 has been renamed to Emgu.CV.windows.runtime.cuda.dldt in v4.4.
    • Added Emgu.CV.Bitmap nuget package that contains the conversion code from Emgu CV images to Bitmap. This package targets .NetStandard platform. If you are running on Linux / Mac OS platform, you will need to make sure native GDI+ is installed in your system, or it will failed on dependency when Bitmap is used.
    • Added Emgu.CV.UI nuget package. This package contains Emgu.CV.Platform.NetFramework.dll and Emgu.CV.Platform.NetCore.dll originally contained in Emgu.CV nuget pacakge in v4.3. These two dlls are only compatible with Windows platform.
    • Emgu.CV nuget package now only targets .NetStandard platform.
    • Split up Emgu.CV.runtime.cuda.dnn nuget packages into 3: Emgu.CV.runtime.cuda.dnn.cnn, Emgu.CV.runtime.cuda.dnn.ops & Emgu.CV.runtime.cuda.dnn. No changes is needed on the user side, the dependency resolution will be handled automatically by nuget package manager.
  • The 64-bit commercial binary for Windows is compiled with Intel Inference Engine backend v2020.4 (instead of v2020.3 in the v4.3 release)
  • Enable IPP and CPU dispatch for windows commercial build.
  • For Android runtime, in v4.3 we switched to from Open CV's custom toolchain to the android toolchain from NDK. There is a bug in Android NDK that enabled debug symbols even in release built. As a result, the Android binary is ~6 times larger. We have updated our build flags to force to exclude the debug symbols in release built. In v4.4, the android binary size are back to the v4.2 level.
  • 4.4.0 release is build with Tesseract 4.1.1
  • GPU release is build with CUDA 11.0 instead of CUDA 10
  • Added Intensity Transform module
  • Added Rapid module
  • Added eigen3 submodule to build alphamat on windows and android.
  • Enable saliency module for Android.
  • Re-enable CudaCascadeClassifier class.
  • Enable building DynamicImage.Asp.NetCore project on Linux, Mac and Raspbian.
  • Added HelloWorld.UWP project for UWP build.
  • Added CvInvoke.ScaleAdd function.
  • Added build script for Ubuntu 20.04
  • VideoCapture.Read function now return bool instead of void.
  • Updated XamarinForms.WPF to base on .NetCore instead of .Net Framework 4.7
  • Updated WPF demo project to base on .NetCore instead of .Net Framework 4.6.1
  • Change DnnInvoke.GetAvailableBackends() to property DnnInvoke.AvailableBackends
  • Moved SIFT from XFeatures2D to Features2D namespace
  • VideoCapture.Run function use Emgu.Util.ExceptionHandler instead of System.ServiceModel.Dispatcher.ExceptionHandler.
  • The License Plate detection demo now use DNN with OpenVino model zoo for improved detection accuracy.


Enhancements

  • Implemented CI build on github.

Bug Fixes

  • This version can run on workstation without AVX2 instruction set. Fixed the known issue for Open CV 4.3 where the dldt build is not able to run on workstations without AVX2.

Dependency

  • Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack. e.g. for windows server 2012 R2, following this instruction

Emgu.CV-4.3.0

Online Documentation

Browse

Change Log

  • 4.3.0 is based on the Open CV 4.3.0 release.
  • For commercial release, we no longer offer binary compiled with Intel C++ compiler. In many case, Visual C++ is producing binaries as efficient as those produced with Intel C++ compiler. In certain situations (e.g. DNN modules running Facial Landmark demo), the binary produce by Visual C++ even run faster than those produced by C++ compiler. We decided to build our 64-bit commercial release without Intel C++, and instead, including Intel Inference Engine for better performance on the DNN module.
  • The 64-bit commercial binary for Windows is compiled with Intel Inference Engine backend v2020.2. This version required a machine with AVX2 instruction set, or else it will crash when DNN module is loaded. See Known Issues. Official Open CV 4.3 release has the exact same bug from Open Vino.
  • 4.3.0 release is build with Tesseract 4.1.1
  • Mac OS binary is now signed using our developer certificate.
  • Android release is now built with NDK r21. Android demo now has a minSdkVersion="24" & targetSdkVersion="29"
  • Added runtime shared project for Raspbian and Ubuntu
  • Ubuntu and Raspbian build script now require dotnet core 3.1 instead of mono
  • Enable building raspbian and Ubuntu nuget pacakge.
  • Removed ZedGraph dependency from Emgu.CV.UI. HistogramBox now use Open CV's plot module to draw histogram instead of using ZedGraph.
  • Removed CvBlob. It has not been updated for 7 years. Please use opencv's tracking module instead.
  • SIFT detector is enabled by default due patents expiration (without requirement of NONFREE build option)
  • Added Emgu.CV.UI.NetCore project which provide a .NetCore compatible implementation of Emgu.CV.UI
  • Added Emgu.CV.Cuda.NvidiaOpticalFlow_1_0. Only support NVidia Graphic cards with Turing architecture (e.g. RTX 20xx)or above.
  • Added SetWindowProperty, GetWindowProperty & SetWindowTitle to CvInvoke. Renamed NamedWindowType to WindowFlags.
  • Added BundleAdjuster, ExposureCompensator, SeamFinder, Estimator and FeaturesMatcher for stitching.
  • Added CylindricalWarper, CylindricalWarperGpu, PlaneWarperGpu & SphericalWarperGpu for stitching.
  • Added RotationWarper for stitching.
  • Added CvInboke.EstimateChessboardSharpness
  • Added warpImage, estimateTransformation & applyTransformation function for IShapeTransformer
  • Added h/vconcate overload that accept multiple Mats.
  • Added LBPHFaceRecognizer.Histograms property.
  • Rename HoughTypes to HoughModes. Added CvEnum.HoughMode.GradientAlt.
  • Added default value for FeatherBlender's "sharpness" parameter.
  • Added "Values" property for VectorOfObject debugger visualizer, when it is not a VectorOfVector.
  • Turned HelloWorld demo from .NetFramework console app to .NetCore console app.
  • Added BuildInfo.NetCore.Console demo project.
  • Use Xamarin.Plugin.FilePicker to pick file on Xamarin Forms Windows WPF demo.
  • Use MaskRcnn with DNN module for TrafficSignRecognition demo.
  • Apply Non-maximum suppression on MaskRcnn demo.
  • Updated VideoSurveillance demo to use Tracking module instead of CvBlob.
  • Replaced WebDynamicImage demo with DynamicImage.Asp.NetCore demo.
  • Moved ShapeDetection demo to Xamarin Forms.
  • Emgu.Util.Platform can now detect iOS platform.
  • Removed lots of __IOS__ conditional compilation flags in Emgu CV source code.
  • Use .net core 3.1 for unit test.

Enhancements

  • Added Yolo demo project in Xamarin Forms.
  • Demo projects now use async function calls to download model files from Internet.
  • Xamarin forms Android demo (Face Detection, Face Landmark Detection, Mask Rcnn Detection, Stop Sign detection, Yolo) can now use Camera in real time.

Bug Fixes

  • Fixed a bug in WText Dispose function.
  • Fixed a bug in cveSuperresFrameSourceRelease PInvoke.
  • Fixed cveSuperResolutionRelease header definition.
  • Fixed a bug in QuasiDenseStereo.Process implementation.
  • Fixed the spelling on cvePlaneWarperCreate.
  • Fixed a bug in loading image file into Image<Bgr, double>.

Dependency

  • Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack. e.g. for windows server 2012 R2, following this instruction

Known Issues

  • Debugger Visualizer does not work in this release. Getting debugger visualizer working for .NetStandard profile is a much challenging task than we originally thought. We will try to address that in the next release.
  • For commercial release, it is build with Intel Inference Engine backend for Windows. Due to a bug in OpenVINO, the application will crashed on Machines without AVX2 instructions sets. This bug should be fixed in OpenVINO 2020.4 according to the ticket.

Emgu.CV-4.2.0

Online Documentation

Browse

Change Log

  • 4.2.0 is based on the Open CV 4.2.0 release
  • The Emgu.CV dll is now targeting .NetStandard 2.0 instead of .Net 3.5
  • The Emgu.CV.UI dll is now targeting .Net 4.6.1 instead of .Net 3.5
  • The nuget package structures has been changed. Emgu.CV nuget package < 4.2.0 contains both the .Net 3.5 dll and the native dll for Windows. Starting from 4.2.0, Emgu.CV nuget package will contain only the .NetStandard 2.0 profile dll. The native dll comes in separate packages for different platforms. e.g. The Emgu.CV.runtime.windows package contains the native dll for Windows as well as the Emgu.CV.UI dll for Windows when targeting .netframework 4.6.1+. The Emgu.CV-CUDA nuget package has been replaced with Emgu.CV.runtime.windows.cuda nuget package.
  • Open CV CUDA DNN module required Compute 5.3 and higher. Our cuda build not longer build with cuda compute bin option < 5.3.
  • Conversion between Mat, UMat, GpuMat and Image<,> and Bitmap objects requires code changes. The constructors of the Mat, UMat, GpuMat and Image<,> that accepts Bitmap has been removed. It has been added to the Emgu.CV.UI.dll file as an extension method. The dll is part of the Emgu.CV.runtime.windows (or the similar Emgu.CV.runtime.windows.cuda) nuget pacakge. Prior to v4.2.0 this code can be use to create a Mat from Bitmap
    Mat m = new Mat(bitmap);
    . In the 4.2.0 release it has to be written like this
    Mat m = bitmap.ToMat()
    For
    Image<,>.Bitmap
    property in 4.1.x and older release, its equivalent in v4.2.0 is
    Image<,>.AsBitmap()

Enhancements

  • Added DnnSuperres module.
  • Added the turbo color map.
  • Added Vulkan and Cuda backend enumeration for DNN.
  • Added FSR Inpainting type.
  • Added Features2DToolbox.DrawMatches overload.
  • Added CvInvoke.DrawMarker

Bug Fixes

  • Fixed the ambiguous reference compilation error between 'Emgu.CV.Structure.Range' and 'System.Range' to make it compatible with the latest version of Visual Studio

Dependency

  • Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack.

Emgu.CV-4.1.1

Online Documentation

Browse

Change Log

  • nuget packages is now provided for Android and iOS in the commercial release.
  • 4.1.1 is based on the Open CV 4.1.1 release
  • 4.1.1 Windows non-commercial release is built with Visual Studio 2019. 4.1.1 Windows commercial release is built with Intel Compiler and Visual Studio 2017. CUDA release is built with CUDA 10.1
  • Mac OS build is now compatible with Visual Studio for Mac 8.2.3 (build 16) or above
  • iOS and Android demo build is now compatible with Visual Studio Aug 2019 releases or above

Enhancements

  • Added Emgu.CV.PpfMatch3d.ICP class from surface matching module.
  • Added XFeatures2DInvoke.MatchGMS
  • Added LineIterator.
  • Replaced IImage with IInputArray
  • Added Dnn.Net.DumpToFile and Dnn.Net.Dump.
  • Added AP3P, IPPE and IPPESquare to SolvePnpMethod, added more solve PnP functions.

Bug Fixes

Dependency

  • Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack.

Emgu.CV-4.1.0

Online Documentation

Browse

Change Log

  • 4.1.0 is based on the Open CV 4.1.0 release
  • 4.1.0 Windows release is built with Visual Studio 2017. CUDA release is built with CUDA 10.1
  • iOS binary is build with XCode 10.1.
  • Removed Emgu.CV.Contrib.LineDescriptor module. It is no long supported by Open CV due to license issue.

Enhancements

  • Added Matplotlib Perceptually Uniform Sequential colormaps
  • Add Hand-Eye calibration methods
  • Added the Quality module.
  • Added Stereo module and QuasiDenseStereo class.
  • Added RLOFOpticalFlowParameter.
  • Added DenseRLOFOpticalFlow and SparseRLOFOpticalFlow.
  • Disabled Freetype module for iOS. Freetype is disabled by default in Open CV for iOS target.
  • Added CvInvoke.SelectROIs & CvInvoke.SelectROI
  • Added MultiTracker.GetObjects.
  • Added DnnInvoke.GetAvailableBackends

Bug Fixes

  • For Unity, fixed Android ILCPP backend build for arm64-v8a architecture.
  • Fixed a bug in the DPM constructor.

Dependency

  • Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack.

Emgu.CV-4.0.1

Online Documentation

Browse

Change Log

  • 4.0.1 is based on the Open CV 4.0.1 release
  • 4.0.1 Windows release is built with Visual Studio 2017. CUDA release is built with CUDA 10.0
  • The Windows Commercial license will also include the UWP release for ARM architecture in v4.x. We have merged the v3.x Windows Phone license and Windows license into v4.x windows commercial license for simplicity.
  • Tesseract OCR has been updated to 4.0 final release.
  • iOS binary is build with XCode 10.1.
  • Android build no longer support armeabi architecture. It is obsoleted.
  • For Unity, iOS has JPEG disable to avoid conflict with Unity's build-in libjpeg.
  • For Unity, Android build has added arm64-v8a binary support.
  • For Unity, Emgu CV can now compatible with the ILCPP back end for Unity UWP project.
  • The installer can now install Debugger Visualizer for VS2019

Enhancements

  • Added Freetype module
  • Added hfs module from opencv_contrib
  • Added DNN sample for Xamarin Forms WPF.
  • Update Stitcher API to match Open CV 4.0 API
  • Added Decode to QRCodeDetector.
  • Added EstimateAffine2D & EstimateAffinePartial2D
  • Added BasicFaceRecognizer class.
  • Wrap more functions for FileNode and IAlgorithm.
  • Added CalibrateCameraCharuco
  • Added DrawPlanarBoard, EstimatePoseBoard and GetBoardObjectAndImagePoints to ArucoInvoke.
  • Added KNearest.FindNearest
  • Added Relu and LeakyRelu activation type to ANN_MLP.
  • Added the RNG class.
  • Added GetData for UMat.
  • Added operator overload to Mat class.
  • Added operator overload to UMat.
  • Update the EstimateTransform function to match opencv implementation.
  • Added OcrInvoke.SetLocale function to set locale for tesseract.
  • Added FaceLandmarkDetectionPage demo page.
  • Allow for picking images on Xamarin Forms WPF project.
  • Implement ROI copying for Image<,> class without changing the IplImage.ROI property.
  • Added CudaConvolution.
  • Added CudaInvoke.DrawColorDisp
  • Added RetinaFastToneMapping
  • Added a Debugger and the GetData function for GpuMat.
  • Added CvInvoke.SetGlDevice
  • Updated CvEnum.CapProp
  • Added VideoCapture.BackendName

Bug Fixes

  • Fixed issue #220: Bug in CudaSparsePyrLKOpticalFlow
  • Fixed issue #210: CvException.ErrorStr crashes program. Thanks to Jeremy
  • Fixed issue #181
  • Fixed issue #173
  • Fix a bug in the PInvoke signature for DisparityBilateralFilterCreate.
  • Fixed a bug in CudaCannyEdgeDetector, CudaHOG constructor and DPM.
  • Fixed a bug in Mat.GetData when there are multiple channels.
  • Fix a bug in FitEllipseDirect.
  • Change the Emgu.CV.World project output path such that a rebuild will not clean out the "x86" or "x64" folder.
  • Added the missing cudaSparsePyrLKOpticalFlowRelease implementation.

Dependency

  • Note that for Windows platform, this release is build with MSMF. MSMF is pre-installed for Windows 10 Home or Windows 10 Pro releases. For Windows N or certain Windows Server releases, it may require you to download and install Windows Media Feature pack.

Version History 3.x Archive

Version History 3.x

Version History 2.x Archive

Version History 2.x

Version History 1.x Archive

Version History 1.x