v0.6.0

New features:

• Update to OpenCV 4.5.2
• Update to CUDA 11/ CUDNN 8
• Add 2 events for plugins execution: workflowStarted and workflowFinished
• Big Ikomia API refactoring

Improvements:

• Plugin import strategy
• Start MLFlow and Tensorboard servers only if not already started

Bug fixes:

• missing GIL protection
• fix visual artefacts when using line break in plugin description
• ...

v0.5.0

New features:

• Tensorboard integration
• Export image + graphics in dataset view
• Access code folder or repository directly from plugin documentation
• Embed graphics in processed video while exporting
• Plugin documentation can be extended as HTML or Markdown file

Improvements:

• Plugins are loaded even if user is not connected
• Add possibility to select export format of workflow outputs
• Graphics layers are no longer modifiable in output view
• Task derived from Python TrainProcess class include a native dict to handle parameters
• Same colormap for all images of a dataset

Bug fixes:

• Major memory leak in Python bindings
• Version number update for plugins
• Python GIL management for mixed type objects destructors (C++/Python)
• Allow image visualization during batch processing
• Video export on Windows platform
• ...

v0.4.1

Bug fixes release

Bug fixes:

• Wrong folder name for plugins installed from the Marketplace (plugins then fail to load in Ikomia Studio)
• Invalid connection between project folder of classification dataset and training algorithms for classification (ResNetTrain, ResNeXtTrain, MnasNetTrain)
• Remove duplicate task execution on complex workflows
• Python initialization issue
• PyTorch issue with std::regex
• Wrong progress bar step count for workflows with self-input task

v0.4.0

New features:

• Add "displayable" option to workflow outputs
• Add new workflow input/output: multidimensional array

Improvements:

• Display optimization for dataset loader tasks
• Better time image sequence management
• Add possibility to customize I/O description from Python API
• Installation of Python plugin dependencies in a separate thread
• Batch mode: reshape auto-save folder structure
• Better compatibility check for C++ and Python API
• Update to MLflow 1.14
• Update to Numpy 1.19.0

Bug fixes:

• Issue #1
• Stderr/stdout redirection (Windows)
• Wrong order in multi-steps Python requirements installation
• Wrong path to Python Scripts folder for MLflow (Windows)
• Runs creation in MLflow
• ...

v0.3.0

New features:

• OpenCV 4.5.0
• MLflow integration for Deep Learning training algorithms.
• Deep Learning dataset display
• Batch processing mode
• Add OpenCV SIFT process
• Open images from a folder (recursive)
• Display input/output description in the workflow editor
• Allow task with no input
• Multi-steps installation of dependencies for Python plugins: create requirements1.txt, requirements2.txt... to ensure sequential installation.

Improvements:

• Reshape Python API structure
• Video management
• Notification center display
• OpenCV Hough process

Bug fixes:

• Display task output information in the workflow editor
• Reshape image list as volume
• User connection check
• Tasks run order in several workflow configuration
• Memory leak in Python wrapper
• ...