Evo 7.12.0-TM26061101
Evo LPR Engine
Loading...
Searching...
No Matches
Release Notes

Runtime Libraries

7.12.0

Executive Summary

  1. Added management for drawing search area's boundary on live view of the FAVEngine.
  2. Added several API functions in C++, C, C# and Python to support the functionality added above.

Backward-Incompatible API Changes

  • C++
    • Additions
      1. the method 'Evo::FAVEngine::SetColorOfSABL()'.
      2. the method 'Evo::FAVEngine::SetDepthOfSABL()'.
      3. the method 'Evo::FAVEngine::DrawSABL()'.
  • C
    • Additions
      1. the function 'EvoFAVE_SetColorOfSABL()'.
      2. the function 'EvoFAVE_SetDepthOfSABL()'.
      3. the function 'EvoFAVE_DrawSABL()'.
  • C#
    • Additions
      1. the method 'Evo.FAVEngine.SetColorOfSABL()'.
      2. the method 'Evo.FAVEngine.SetDepthOfSABL()'.
      3. the method 'Evo.FAVEngine.DrawSABL()'.
  • Python
    • Additions
      1. the method 'PyEvo.FAVEngine.set_sabl_color()'.
      2. the method 'PyEvo.FAVEngine.set_sabl_depth()'.
      3. the method 'PyEvo.FAVEngine.draw_sabl()'.

7.11.0

Executive Summary

  1. Added an engine parameter for minimal search of the FAVEngine.
  2. Added several API functions in C, C# and Python bindings to support the added engine parameter above.

Backward-Incompatible API Changes

  • C++
    • Additions
      1. the enumerator 'EvoParam::EVOPARAM_MIN_SEARCH'.
  • C
    • Additions
      1. the function 'EvoEng_SetParam_MinSearch()'.
      2. the function 'EvoEng_GetParam_MinSearch()'.
  • C#
    • Additions
      1. the method 'Evo.Engine.SetParamMinSearch()'.
      2. the method 'Evo.Engine.GetParamMinSearch()'.
  • Python
    • Additions
      1. the method 'PyEvo.Engine.set_param_min_search()'.
      2. the method 'PyEvo.Engine.get_param_min_search()'.

7.10.0

Executive Summary

  1. Enhanced stability and compatibility of the module 'VideoIO'.
  2. Added an error code related to video streaming.

7.9.0

Executive Summary

  1. Optimized some parameters for inputs to the LPBQD model.

7.8.0

Executive Summary

  1. Applied Intel's OpenVINO 2025.4 and updated other most external libraries.
  2. Added information for the model precision to the 'GTMC'.
  3. Disabled support for the DNN device 'NPU'.
  4. Enhanced basic configuration for each DNN device.

7.7.0

Executive Summary

  1. Applied Intel's OpenVINO 2024.0

7.6.0

Executive Summary

  1. Applied Intel's OpenVINO 2023.1

7.5.0

Executive Summary

  1. Applied Intel's OpenVINO 2023.0.1
  2. Optimized scaling LPD's input images.

Bug Fixes

  1. Fixed calculating support expiration of the license.

7.4.0

Executive Summary

  1. Applied Intel's OpenVINO 2023.0
  2. Removed the DNN devices which are no more supported.
  3. Removed the trained model files with half precision.
  4. Deprecated the concept 'DNN Model Data Format'.
  5. Introduced 'Storage Formats' and 'Runtime Modes' in the DNN Device Descriptor.
  6. Introduced the automatic selection of DNN model's execution precision.
  7. Optimized scheduling inference threads on multi-core processors.
  8. Supprted multiple GPUs.
  9. Improved stability.

Backward-Incompatible API Changes

  • C++
    • Additions
      1. Added the method 'Evo::DDI::GetFullName()'.
    • Deprecations
      1. Deprecated the method 'Evo::DDI::GetNumDMDF()'.
      2. Deprecated the method 'Evo::DDI::GetDMDF()'.
  • C
    • Additions
      1. Added the function 'EvoDDI_GetFullName()'.
    • Deprecations
      1. Deprecated the function 'EvoDDI_GetNumDMDF()'.
      2. Deprecated the function 'EvoDDI_GetDMDF()'.
  • C#
    • Additions
      1. Added the method 'Evo.DDI.GetFullName()'.
    • Deprecations
      1. Deprecated the method 'Evo.DDI.GetNumDMDF()'
      2. Deprecated the method 'Evo.DDI.GetDMDF()'
  • Python
    • Additions
      1. Added the method 'DDI.get()'.
    • Deprecations
      1. Deprecated the method 'DDI.gather()'.

7.3.0

Executive Summary

  1. Sophisticated the license policy.

7.2.0

Executive Summary

  1. Applied Intel's OpenVINO 2022.3
  2. Improved performance in the throughput run-mode.
  3. Improved stability.

7.1.0

Executive Summary

  1. Improved performance of the LPC model.
  2. Supported variable output size in the fisheye-lens model.
  3. Improved logic of DNN model loading.

Backward-Incompatible API Changes

  • C++
    • Additions
      1. Added the method 'Evo::CC::Model::GetMaxResol()'.
      2. Added the methods 'Evo::CC::Fisheye::GetOutputSize()' and 'Evo::CC::Fisheye::SetOutputSize()'.
  • C
    • Additions
      1. All corresponding to C++
  • C#
    • Additions
      1. All corresponding to C++
  • Python
    • Additions
      1. All corresponding to C++

7.0.0 Beta

Executive Summary

  1. Introduced new APIs which are more compact and developer-friendly.
  2. Supported correction of fisheye lens distortion.
  3. Removed the functionality, Automobile Front Face Classification.
  4. Supported input images of noncontinuous planar pixel formats.
  5. Applied license V2 which is more subdivided.
  6. Applied version 2022.2 of Intel's OpenVINO runtime.
  7. Improved stability.

Bug Fixes

  1. Wrong initialization about empty DDD in the class 'TSSEngine'.

Backward-Incompatible API Changes

  • C++
    • Additions
      1. Added several new enumerators to the 'EvoRC'.
      2. Created the camera-calibration related classes 'Evo::CC::Model', 'Evo::CC::Fisheye'.
      3. Added the library 'EvoCC
      4. Created methods for item selection in the classes 'Evo::DDI', 'Evo::LPI', 'Evo::GOP'.
    • Modifications
      1. Renamed majority of enumerators, namespaces and methods.
      2. Replaced the structure 'EvoRawImg' by 'EvoImage'.
      3. Renamed the libraries 'EvoGenICam.dll'/'libEvoGenICam.so' to 'EvoGIC.dll'/'libEvoGIC.so'.
    • Removals
      1. Removed all the deprecated ones.
      2. Removed all about the functionality 'Automobile Front Face Clacification'.
      3. Removed all the staic methods to free or destroy objects in each class.
      4. Removed the argument 'ccodes' in 'Evo::Initialize()'.
      5. Removed the enumerator 'EVOPARAM_SCRIPT_CONF_THRES' and its related functionality.
  • C
    • Additions
      1. All corresponding to C++
    • Modifications
      1. All corresponding to C++
      2. Replaced all the handles and contextes by object-IDs.
    • Removals
      1. All corresponding to C++
  • C#
    • Additions
      1. All corresponding to C++
    • Modifications
      1. All corresponding to C++
      2. Replaced all the handles and contextes by object-IDs.
    • Removals
      1. All corresponding to C++
  • Python
    • Additions
      1. All corresponding to C++
    • Modifications
      1. All corresponding to C++
    • Removals
      1. All corresponding to C++

6.11.1

Bug Fixes

  1. Fixed error related to checking USB license under Windows platform.

6.11.0

Executive Summary

  1. Changed license policy.

6.10.0

Executive Summary

  1. Changed to apply ver 0.6.0 of GLog.
  2. Simplified code structure related to log.
  3. Added checking log initialization.

6.9.0

Executive Summary

  1. In the FAVEngine, enhanced object tracking.

6.8.0

Executive Summary

  1. In the LPI, deprecated GetAmHeadDir() and GetAmType().
  2. In the Output parameter, deprecated 'enAmType'.
  3. In the FAVEngine, invalidated GetPos() in LPI.
  4. In the FAVEngine, enhanced object tracking.
  5. In the FAVEngine, changed the origin from video's center to LP's center calculating the heading direction.
  6. In the FaVEngine, corrected the images without LPs were saved.

Backward-Incompatible API Changes

  • C++
    • Added
      1. Evo::GOP::GetPos()
      2. Evo::GOP::GetHeadDir()
  • C
    • Added
      1. EvoGOP_GetPos()
      2. EvoGOP_GetHeadDir()
  • C#
    • Added
      1. All corresponding to C
  • Python
    • Added
      1. All corresponding to C

6.7.0

Executive Summary

  1. Merged the module 'EvoLicense.dll' into other module and removed it.
  2. In the FAVEngine, enhanced object tracking using different window sizes and geometric translation.

Bug Fixes

  1. In the 'EvoImgData::IsValid()', removed checking alignment of stride.

6.6.0

Executive Summary

  1. In the FAVEngine, improved saving image files.
  2. Supported YCbCr color space in the live view.
  3. Replaced the structure 'EvoRawImg' with more flexible 'EvoImgData' for internal usage.

Bug Fixes

  1. In Linux platform, fixed the problem that live view is not displayed.
  2. In Python API, fixed the exception raised at Python 3.8 or higher.

6.5.0

Executive Summary

  1. Applied Intel's OpenVINO 2022.1
  2. made logic of each model compact.
  3. Improved the performance and stability.

6.4.0

Executive Summary

  1. In the FAVEngine, changed several parts of tracking algorithm.

Bug Fixes

  1. In the FAVEngine, handled OpenCV related exceptions.

6.3.0

Executive Summary

  1. In the FAVEngine, improved entire tracking process.

Bug Fixes

  1. In the FAVEngine, initialized the prediction estimates in Kalman Filter.

6.2.0

Executive Summary

  1. Improved RTSP compatability.

6.1.0

Executive Summary

  1. Supported Intel's 12th generation Core processors in the FAVEngine.
  2. Supported I420, YV12 in available color spaces of the input image.

6.0.0

Executive Summary

  1. Determined the unreliable characters by LPSC considering the result of LPSD.

Bug Fixes

  1. In the Python API, restored the length argument when the C API 'EvoLPI_GetCharConf()' called.

6.0.0 RC3

Executive Summary

  1. On Windows platform, upgraded the build toolset to MSVC-2022.
  2. Updated the Boost library to 1.78.0
  3. Updated the GLEW library to 2.2.0
  4. Updated the libjpeg-turbo library to 2.1.2
  5. Optimized trigger process in the TSSEngine.
  6. Optimized deinitialization process in the TSSEngine.

Bug Fixes

  1. In the TSSEngine, added missing initialization check.
  2. In the FAVEngine, removed deadlock possibility.

6.0.0 RC2

Executive Summary

  1. Supported Python 3.8 and higher.

6.0.0 RC1

Executive Summary

  1. Supported GenICam-compatible cameras.
  2. Intel's OpenVINO library updated to 2021.4.752
  3. On Linux platform, Intel's MediaSDK library updated to 21.3.5
  4. Improved system stability.

Backward-Incompatible API Changes

  • C++
    • Added
      1. 'EvoGenICam.lib' import library for GenICam.
      2. Error codes related to GenICam.
      3. 'Evo::GC::Admin' and 'Evo::GC::Dev' classes.
      4. 'Picture.hpp' and 'GenICam.hpp' header files.
      5. NV12 and NV21 pixel formats.
      6. In 'EvoRawImg' structure, constructors overloaded and converting operator added.
    • Modified
      1. Some error codes for general/specific purpose.
      2. Base integer numbers of 'EvoRC' codes.
      3. 'Evo::Picture' and 'Evo::GOP' classes moved into the header file 'Picture.hpp'.
      4. 'Evo::DDI::Destroy() renamed to 'Evo::DDI::Free()'.
      5. 'Evo::LPI::Destroy() moved to 'Evo::Engine::FreeLPI()'.
      6. 'Evo::Picture::Destroy()' moved to 'Evo::TSSEngine::FreePic()'.
      7. 'Evo::GOP::Destroy()' moved to 'Evo::FAVEngine::FreeGOP()'.
      8. The arguments of 'Evo::SSEngine::Init()'.
      9. 'Evo::TSSEngine::Init()' overloaded for IP-Protocol and GenICam.
      10. 'Evo::FAVEngine::Init()' overloaded for IP-Protocol and GenICam.
      11. Arguments and return value of 'Evo::FAVEngine::GetNumLPIs()'.
    • Removed
      1. 'Evo::TSSEngine::Capture()' method.
    • Deprecated
      1. Output engine parameter for automobile type and its confidence.
  • C
    • Added
      1. 'EvoGC_AdminHandle' and 'EvoGC_DevHandle' structures for GenICam.
      2. 'EvoGC_Admin_*' and 'EvoGC_Dev_*' functions for GenICam.
      3. 'EvoTSSE_Init_IPP()' function.
      4. 'EvoTSSE_Init_GC()' function.
      5. 'EvoFAVE_Init_IPP()' function.
      6. 'EvoFAVE_Init_GC()' function.
    • Modified
      1. 'EvoDDI_FreeCtx()' function renamed to 'EvoDDI_Free()'.
      2. 'EvoLPI_FreeCtx()' function renamed to 'EvoEng_FreeLPI()'.
      3. 'Evo_GetParam_*' and 'Evo_SetParam_*' functions renamed to 'EvoEng_GetParam_*' and 'EvoEng_SetParam_*'.
      4. 'EvoPic_FreeCtx()' function renamed to 'EvoTSSE_FreePic()'.
      5. 'EvoGOP_FreeCtx()' function renamed to 'EvoFAVE_FreeGOP()'.
      6. 'EvoSSE_Alloc()' function renamed to 'EvoSSE_Create()'.
      7. 'EvoSSE_FreeHandle()' function renamed to 'EvoSSE_Destroy()'.
      8. 'EvoTSSE_Alloc()' function renamed to 'EvoTSSE_Create()'.
      9. 'EvoTSSE_FreeHandle()' function renamed to 'EvoTSSE_Destroy()'.
      10. 'EvoFAVE_Alloc()' function renamed to 'EvoFAVE_Create()'.
      11. 'EvoFAVE_FreeHandle()' function renamed to 'EvoFAVE_Destroy()'.
    • Removed
      1. 'EvoTSSE_Capture()' function.
      2. 'EvoTSSE_Init()' function.
      3. 'EvoFAVE_Init()' function.
  • C#
    • Added
      1. All corresponding to C
    • Modified
      1. All corresponding to C
  • Python
    • Added
      1. All corresponding to C
    • Modified
      1. All corresponding to C

6.0.0 Beta-3

Executive Summary

  1. Improved algorithm to find the best candidates in the FAVEngine.
  2. Improved FAVEngine performance using parallel processing.
  3. Supported asynchronous inference of the LPD model.

6.0.0 Beta-2

Executive Summary

  1. Improved algorithm to find the best candidates in the FAVEngine.
  2. Added optional creation of optimal DDDs in FAVEngine.

6.0.0 Beta-1

Executive Summary

  1. Supported simultaneous multiple countries.
  2. Supported Korea's Military Opreation Vehicles.
  3. Improved stability.

Bug Fixes

  1. In some example applications, fixed wrong API usages.

Backward-Incompatible API Changes

  • C++
    • Added
      1. Error codes related to multiple countries.
    • Modified
      1. 'Evo::Initialize()' function.
    • Removed
      1. 'EvoCC' enumeration.
      2. 'Evo::GetVerStr()' function.
  • C
    • Added
      1. All corresponding to C++
    • Modified
      1. All corresponding to C++
    • Removed
      1. All corresponding to C++
  • C#
    • Added
      1. All corresponding to C++
    • Modified
      1. All corresponding to C++
  • Python
    • Added
      1. All corresponding to C++
    • Modified
      1. All corresponding to C++

5.9.0

Executive Summary

  1. For the LPD model, applied Gaussian Pyramid to preprocessing to improve downscaling's accuracy.
  2. In the FAVEngine, removed recycling image preprocssing.
  3. In the FAVEngine, changed comparison of the confidences to update the best detection.

5.8.0

Executive Summary

  1. Recycled LPD's preprocessed input image to FAVEngine to increase performance.

5.7.0

Executive Summary

  1. Changed some codes to apply version 2021.4 of OpenVINO library.
  2. Enabled the local caches inherent in OpenCV and OpenVINO libraries.

5.6.1

Bug Fixes

  1. In the FAVEngine, fixed that deinitialization is blocked while waiting for the output queue.
  2. In the C APIs, fixed inconsistency of the return codes and related problems.

5.6.0

Executive Summary

  1. Added OpenCL support in preprocessing the input images.
  2. For selection of OpenCV devices, added the environment variable 'EVOENG_OCL_DEV'.
  3. For LPD's accuracy, added preprocessing of input images.
  4. Improved engine's stability.

Bug Fixes

  1. In case of using iGPUs as DNN device, fixed bugs of OpenVINO's version 2021.3

5.5.0

Executive Summary

  1. For the FAVEngine on Windows. added the funtionality to search existing h/w CODECs for Intel's QSV and make it available for decoding if found.

Bug Fixes

  1. In the FAVEngine, supported both of tcp and udp in RTSP.

5.4.0

Executive Summary

  1. Modified the module 'EvoVideoIO' to update FFmpeg and Intel's MediaSDK libraries for the FAVEngine.
  2. Optimized searching vertices on LP's bounding quardrangle.

5.3.0

Executive Summary

  1. Added API to get character confidences for some languages.

Backward-Incompatible API Changes

  • C
  • C#
    • Added
      1. All corresponding to C
  • Python
    • Added
      1. All corresponding to C

5.2.0

Executive Summary

  1. Optimized the script recognition logic.

5.1.0

Executive Summary

  1. Supported Thailand's license plates.
  2. Optimized the CSM library.

5.0.0

Executive Summary

  1. Intel's OpenVINO library updated to 2021.3.394
  2. DNN Framework changed from Caffe to PyTorch.
  3. Retrained all the DNN models using PyTorch.
  4. Reliability of LPC results improved.

5.0.0 Beta

Executive Summary

  1. Added Fully Automatic Video Engine.
  2. Retrained all the DNN models.
  3. Improved stability.
  4. Updated some external libraries.

Backward-Incompatible API Changes

  • C++
    • Added
      1. 'Evo::FAVEngine' class for Fully Automatic Video Engine.
      2. 'Evo::GOP' class for group of pictures.
      3. 'Evo::Picture::GetTime()' method.
      4. 'Evo::LPI::GetAmHeadDir()' method.
      5. 'enAmHeadDir' member in 'EvoOutput' structure.
      6. 'EvoTimeUnit' enumeration.
      7. 'EvoHeadDir' enueration.
      8. Several parameters in 'EvoParam' for Fully Automatic Video Engine.
      9. Several EvoRC codes.
    • Modified
      1. 'Evo::GetLastRC()' function.
      2. 'Evo::TSSEngine::Trigger()' method.
    • Removed
      1. 'enUnkStr' member in 'EvoOutput' structure.
      2. All the deprecated APIs from 4.x.x
  • C
    • Added
      1. All corresponding to C++
    • Modified
      1. All corresponding to C++
    • Removed
      1. All corresponding to C++
  • C#
    • Added
      1. All corresponding to C++
    • Modified
      1. All corresponding to C++
    • Removed
      1. All corresponding to C++
  • Python
    • Added
      1. All corresponding to C++
    • Modified
      1. All corresponding to C++
    • Removed
      1. All corresponding to C++

4.5.0

Executive Summary

  1. Supported software license for use of NVMe SSD.

Bug Fixes

  1. In case of Linux version, fixed link error of C-API library.

4.4.0

Executive Summary

  1. Updated Intel's OpenVINO library.
  2. Changed the string format of the runtime version.
  3. Improved initialization sequence of Triggered Snapshot Engine.

Bug Fixes

  1. Fixed invisible live-view.
  2. Removed properties which make video streaming incompatable.

4.3.0

Executive Summary

  1. Retrained all the DNN models.
  2. On Triggered Snapshot Engine, added live-view functionality.
  3. Supported dynamic module loading to reduce library dependency and loading latency.

Backward-Incompatible API Changes

  • C++
    • Added
      1. 'Evo::TSSEngine::OpenLiveView()' method.
      2. 'Evo::TSSEngine::CloseLiveView()' method.
  • C
    • Added
      1. 'EvoTSSE_OpenLiveView()' function.
      2. 'EvoTSSE_CloseLiveView()' function.
  • C#
    • Added
      1. 'Evo.TSSEngine.OpenLiveView()' method
      2. 'Evo.TSSEngine.CloseLiveView()' method
  • Python
    • Added
      1. 'PyEvo.TSSEngine.open_live_view()' method
      2. 'PyEvo.TSSEngine.close_live_view()' method

4.2.0

Executive Summary

  1. Optimized switch logic between DNN models.
  2. Added Triggered Snapshot Engine for network-streamed video input.

Bug Fixes

  1. In the 'EvoEngine.dll', fixed crash by invalid log message.
  2. In the 'EvoLicense.dll', fixed crash which result from COM Win32 APIs.

Backward-Incompatible API Changes

  • C++
    • Added
      1. 'Evo::TSSEngine' and 'Evo::Picture' classes.
  • C
    • Added
      1. 'Evo_SetParam_*()' functions.
      2. 'Evo_GetParam_*()' functions.
      3. 'EvoPic_*()' functions.
      4. 'EvoTSSE_*()' functions.
    • Deprecated
      1. 'EvoSSE_GetParam_*()' functions.
      2. 'EvoSSE_SetParam_*()' functions.
  • C#
    • Added
      1. 'Evo.SSEngine' class.
      2. 'Evo.Picture' class.
      3. 'Evo.Func.GetParam*()' methods.
      4. 'Evo.Func.SetParam*()' methods.
    • Deprecated
      1. 'Evo.SnapshotEngine' class.
  • Python
    • Added
      1. 'PyEvo.Picture' class.
      2. 'PyEvo.TSSEngine' class.
      3. 'PyEvo.set_param_*()' functions.
      4. 'PyEvo.get_param_*()' functions.
    • Deprecated
      1. 'PyEvo.SSEngine.set_param_*()' methods.
      2. 'PyEvo.SSEngine.get_param_*()' methods.

4.1.0

Executive Summary

  1. Improved license plate classification logic considering its detection information.
  2. Improved script recognition logic considering its detection information.
  3. Redesigned DNN model for detecting license plate's quardrangle.
  4. Retrained all the DNN models.
  5. Optimized the default confidence thresholds of some DNN models.
  6. Updated Intel's OpenVINO library to 2021.1.110
  7. In the 'TestSSE' program, added '–img_fmt' and '–search_rect' flags.

Bug Fixes

  1. In the 'TestSSE' program, handled the input images which can't be decoded.
  2. Fixed abnormal termination on Intel's 2Gen Xeon Scalable Processors.

4.0.1

Executive Summary

  1. Improved DNN model for license plate detection. So, it is more reliable and a little faster.
  2. Optimized the default confidence thresholds of some DNN models.
  3. Retrained all the DNN models.

Bug Fixes

  1. Fixed unhandled exception in the license library.
  2. Fixed character transcoding problem in the license library.
  3. Fixed bounding-box overrange in the license plate detection.

4.0.0

Executive Summary

  1. Added 'EVOENG_PUBLIC_DIR' environment variable for Linux platform.
  2. Enabled trial license without the need for certification devices.
  3. In the 'TestSSE' program, added '–no_gui' flag to support terminal on Linux platform.
  4. Retrained some DNN models.
  5. Optimized the default confidences of some DNN models.
  6. For Korea country, added 'ěž„' character.

Bug Fixes

  1. Fixed OpenCL caching problem related to 'cl_cache_dir' environment variable on Linux platform.

4.0.0 RC3

Executive Summary

  1. To prevent name duplications with other external libraries. renamed all the libraries more specifically.
  2. As engine inputs, enabled encoded image data (JPEG, PNG, BMP and so on) in memory buffer.

Backward-Incompatible API Changes

  • C++
    • Renamed
      • Windows
        1. 'SSEngine.lib' -> 'EvoSSE.lib'
        2. 'Engine.lib' -> 'EvoEngine.lib'
      • Linux
        1. 'libSSEngine.so' -> 'libEvoSSE.so'
        2. 'libEngine.so -> 'libEvoEngine.so'
      • Common
        1. 'struct EvoImgBuf' -> 'struct EvoRawImg'.
  • C
    • Renamed
      • Windows
        1. 'C_API.lib' -> 'EvoCAPI.lib'
      • Linux
        1. 'libC_API.so -> 'libEvoCAPI.so'
      • Common
        1. 'struct EvoImgBuf' -> 'struct EvoRawImg'.
        2. 'EvoSSE_Run_IF()' -> 'EvoSSE_Run_ImgFile()'.
        3. 'EvoSSE_Run_IB()' -> 'EvoSSE_Run_RawImg()
  • C#
    • Renamed
      1. 'CS_API.dll' -> 'EvoCSAPI.dll'
      2. 'Evo.ImgBuf' -> 'Evo.RawImage'.
  • Python
    • Renamed
      1. 'PyEvo.ImgBuf' -> 'PyEvo.RawImage'.
      2. 'PyEvo.SSEngine.run_with_imgbuf()' -> 'PyEvo.SSEngine.run_with_rawimg()'.

4.0.0 RC2

Executive Summary

  1. To specify DNN Run Mode (Latency-Oriented or Throughput-Oriented), introduced DNN Run Mode Descriptor (DRMD) in engine's global initialization.

Backward-Incompatible API Changes

  • C++
    1. Changed the arguments of 'Evo::Initialize()' function.
  • C
    1. Changed the arguments of 'Evo_Initialize()' function.
  • C#
    1. Changed the arguments of 'Evo.Func.Initialize()' function.
  • Python
    1. Changed the arguments of 'PyEvo.initialize()' funcion.

4.0.0 RC1

Executive Summary

  1. No more support backward compatibility with 3.x.x
  2. Removed Tesseract OCR and adapted inherent algorithm based on DNN to recognize the scripts to increase accuracy.
  3. Added the confidence threshold to determine whether detected scripts are reliable.

Backward-Incompatible API Changes

  1. Removed all the deprecated APIs from 3.x.x

Trained Models

TM26061101

Executive Summary

  • Korea
    1. Retrained several models with new samples.

TM26041801

Executive Summary

  • Korea
    1. Retrained several models with new samples.

TM26021101

Executive Summary

  • Korea
    1. Fixed the problem that some models can't be loaded on iGPU.

TM26020401

Executive Summary

  • Korea
    1. Prominently enhanced detection of the plate and its scripts.

TM26013001

Executive Summary

  • Korea
    1. Enhanced structures of some models.
    2. Retrained several models with new samples.

TM26011601

Executive Summary

  • Korea
    1. Enhanced all the DNN models for the accuracy.
    2. Retrained all the models with new samples.

TM24120101

Executive Summary

  • Korea
    1. Retrained the sevaral models with new samples.

TM23101301

Executive Summary

  • Korea
    1. Retrained the sevaral models with new samples.

TM23100501

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

TM23090501

Executive Summary

  • Korea
    1. Improved architectures of LPD model for performance and accuracy.
    2. Retrained some models with new samples.

TM23081201

Executive Summary

  • Korea
    1. Improved architectures of LPBQD, LPC, LPSD and LPSC models for performance and accuracy.
    2. Retrained all the models with new samples.
    3. Enhanced recognition of construction vehicles.

TM23070401

Executive Summary

  • Korea
    1. Enhanced architecture of the LPSC model.
    2. Retrained all the models with new samples.
    3. Improved recognition of construction vehicles

TM23062501

Executive Summary

  • Korea
    1. Retrained the LPD and LPBQD models with new samples.

TM23061801

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

TM23050301

Executive Summary

  • Korea
    1. Retrained LPD, LPBQD, LPSD and LPSC models with new samples.

TM23031201

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

TM23020101

Executive Summary

  • Korea
    1. Retrained LPD, LPBQD, LPC and LPSC models with new samples.

TM22122701

Executive Summary

  • Korea
    1. Retrained LPBQD, LPC and LPSD models with new samples.

TM22122001

Executive Summary

  • Korea
    1. Retrained the models of LPD, LPBQD and LPC with new samples.

TM22120501

Executive Summary

  • Korea
    1. Enhanced architectures of the LPBQD, LPC, LPSD and LPSC models.
    2. Retrained all the models with new samples.

TM22112301

Executive Summary

  • Korea
    1. Enhanced architectures of the LPC and LPSC models.
    2. Retrained most of the models with new samples.

TM22110101

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

TM22102801

Executive Summary

  • Korea
    1. Made the LPD model more sophisticated.
    2. Made the LPC model more compact and trained it using improved loss function.
    3. Retrained all the models with new samples.

TM22102001

Executive Summary

  • Korea
    1. Enhanced architecture of the LPC model to increase accuracy.
    2. Retrained all the models with new samples.
    3. Optimized all the models with OpenVINO 2022.2

TM22090701

Executive Summary

  • Korea
    1. Enhanced architecture of LPD model for accuracy.
    2. Retrained a number of models with new samples.

TM22081701

Executive Summary

  • Korea
    1. Retrained a number of models with new samples.

TM22072801

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

TM22071801

Executive Summary

  • Korea
    1. Retrained several models with new samples.

TM22070801

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

TM22062501

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

TM22061501

Executive Summary

  • Korea
    1. In the LPC model, added new type of CV license plate announced in 2021.
    2. Retrained some other models with new samples.

TM22061001

Executive Summary

  • Korea
    1. Optimized and retrained some hyper-parameters of LPD model.
    2. Retrained some other models with new samples.

TM22060801

Executive Summary

  • Korea
    1. Improved architecture of LPD model for accuracy.
    2. Retrained all the models with new samples.
    3. Optimized hyper-parameters for some models.

TM22053101

Executive Summary

  • Korea
    1. Enhanced architecture of LPD model for performance and accuracy.
    2. Retrained all the models with new samples.
    3. Optimized hyper-parameters for some models.

TM22051901

Executive Summary

  • Korea
    1. Retrained some models with new samples.
    2. Optimized some hyper-parameters for the LPD model.
    3. Increased the input size of LPBQD model to enhance accuracy.

TM22051501

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. Optimized some hyper-parameters training the LPC model.

22050201

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. Optimized some hyper-parameters training the LPSC model.

22033101

Executive Summary

  • Korea
    1. Retrained LPD, LPBQD and LPSC models with new samples.
    2. Optimized all the moddels with Intel's OpenVINO 2022.1
  • Korean MOV
    1. Optimized all the moddels with Intel's OpenVINO 2022.1
  • Thailand
    1. Optimized all the moddels with Intel's OpenVINO 2022.1

22032101

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

22030401

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

22022301

Executive Summary

  • Korea
    1. Improved the LPD model to increase performance.
    2. Retrained the LPD, LPBQD, LPC and LPSC models.

22021801

Executive Summary

  • Korea
    1. Modified the LPD model architecture to increase accuracy and performance.
    2. Retrained the LPD model.

22021401

Executive Summary

  • Korea
    1. Modified the LPBQD model architecture to increase accuracy and performance.
    2. Modified the LPD model architecture to increase accuracy.
    3. Retrained all the models with new samples.

22012001

Executive Summary

  • Korea
    1. Applied regularization in LPSC model.
    2. Modified the LPSC model architecture to increase accuracy and performance.
    3. Modified the LPD model architecture to increase accuracy.
    4. Retrained all the models with new samples.

22010601

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. Optimized the margins in the training of LPSC models.
    3. Trained LPD and LPSD models with updated Loss function.

21122301

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. Optimized the margins/scales in the training of LPC and LPSC models.
    3. For LPC and LPSC models, increased the number of layers for accuracy.
    4. Made the data augmentation more sophisticated w.r.t all the models.
    5. For LPSD model, increased size of the feature dimension.

21121301

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. Increased the margins of LPC and LPSC models.
    3. For LPD model, augmented the sample data using geomatric transforms.

21112501

Executive Summary

  • Korea
    1. Retrained LPSC with new samples.
    2. Applied version 2021.4.752 of OpenVINO toolkit.
  • Korea's MOV
    1. Applied version 2021.4.752 of OpenVINO toolkit.
  • Thailand
    1. Applied version 2021.4.752 of OpenVINO toolkit.

21100501

Executive Summary

  • Korea
    1. Retrained some models with new samples.
    2. Changed data augmentation for training the LPD model.

21092301

Executive Summary

  • Korea
    1. Added models for Military Operation Vehicles.
    2. Retrained the LPD model with new samples.

21090101

Executive Summary

  • Korea
    1. Improved LPD model for better accuracy.
    2. Retrained all the models with new samples.

21081201

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. Changed the backbone network in LPD model for better performance.
    3. Used additional data augmentations training LPC and LPSC models.

21080201

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. In LPD model, reduced the frequency of false-positive w.r.t the background.

21072501

Executive Summary

  • Korea
    1. Retrained all the models with new samples.
    2. Changed the backbone network in LPC model for better accuracy.

21070601

Executive Summary

  • Korea
    1. Applied version 2021.4 of OpenVINO toolkit.
  • Thailand
    1. Applied version 2021.4 of OpenVINO toolkit.

21062801

Executive Summary

  • Korea
    1. Retrained some models with new samples.

21062401

Executive Summary

  • Korea
    1. Optimized hyperparameters of some models.
    2. Retrained some models with new samples.
    3. Improved LPD's detection accuracy.

21061801

Executive Summary

  • Korea
    1. Optimized hyperparameters of some models.
    2. Changed the optimizers of some models.
    3. Retrained all the models with new samples.

21060801

Executive Summary

  • Korea
    1. Retrained all the models with new samples.

21053001

Executive Summary

  • Korea
    1. Improved LPBQD and LPSD models and retrained them.
    2. Retrained all the other models with new samples.

21052601

Executive Summary

  • Korea
    1. Improved LPC and LPSC models and retrained them.
  • Thailand
    1. Improved LPSC model and retrained it.