Releases: Ikomia-dev/IkomiaApi
v0.15.0
v0.14.0
New features:
- Output aggregation for video processing workflow
- Handle webm and mkv video format
- Add a save path attribute for all I/O, useful for retrieving aggregated results from video processing
- Add task and workflow method to handle long time initialization separately (model downloading, model compiling)
Improvements:
- Performance improvement when retrieving compatible algos from Ikomia HUB
- Update dependencies: numpy >=1.26.4, mlflow==2.20.1
- Set compact mode for all JSON-based I/O export
- Add JSON export format for all compatible I/O
Bug fixes:
- Add missing load/save functions for CJsonIO
- Conversion Numpy <-> cv::mat (preparing Python 3.12 / 3.13)
v0.13.0
New features:
- Add 3D points for keypoints detection
- Add new factory to instantiate task parameters from name
Improvements:
- Enable parameters initialization at task instantiation (not just after)
- infer_mmlab_pose_estimation: integration of RTMW3D algorithm
- check algorithm file modification to update auto-completion module (#39 )
Bug fixes:
- Deep copy of task parameters objects
v0.12.0
New features:
- Add function to create demo workflow from HUB algorithm
- Save API configuration at runtime (auto-completion ON/OFF, root folder, Tensorboard config, MLFlow config...)
- Feature to enable/disable workflow task for running
- New HUB algorithms:
- infer_ddcolor_colorisation
- infer_owl_v2
- infer_yolo_v9_seg, train_yolo_v9_seg
- infer_depth_anything_v2
- infer_florence_2_caption, infer_florence_2_ocr, infer_florence_2_detection, infer_florence_2_segmentation
- infer_segment_anything_2
- infer_rembg
- infer_flux_1 (schnell)
- infer_finegrain_segmenter
- infer_donut
Improvements:
- Manage EXIF metadata (orientation)
Bug fixes:
- Avoid duplicate log messages between stderr and stdout
- Color mismatch in semantic segmentation tasks #37
- Deep copy of CWorkflowTaskParam object
- Class labels overflow in semantic segmentation IO
v0.11.1
v0.11.0
New features:
- OpenCV 4.8.1
- Add I/O information in JSON export for workflow introspection
- Add feature to expose task parameters at workflow level
- Add feature to expose task outputs at workflow level
- Add RGB <-> RGBA conversion option in ocv_color_conversion algorithm
- Add environment variable (IKOMIA_AUTO_COMPLETION) to set auto-completion ON/OFF
Improvements:
- Manage log status (info, warning, error) for Python
- Manage RGBA source image
- Indicate Python version in workflow JSON file
- Non string values can be passed to Workflow::set_parameters()
- Forward exception callstack when algorithm failed to load (#33)
- Avoid double algorithm instantiation while loading workflow
Bug fixes:
- Kernel size limit for StackBlur and BoxFilter algorithms
- Handle empty mask for semantic segmentation output
- Register CDatasetIO in I/O factory
- Handle alpha channel for base 64 image conversion
- Circular imports
- ...
v0.10.0
This release marks the transition to the new Ikomia Scale platform. This platform includes a new version of Ikomia HUB with more insights on how to use every algorithms with this API.
New features:
- Switch to new Ikomia HUB from Ikomia Scale platform
- Add possibility to manage public and private HUBs
- Token-based authentication for private HUB
Improvements:
- Add polygons output for instance and semantic segmentation algorithms
- Add methods in CImageIO class to get visual image from other I/O (graphics, object detection, segmentation...)
- Improve speed of auto-completion system (ik module)
Bug fixes:
- Export segmentation masks in PNG format for JSON serialization
- Missing Python binding for CInstanceSegIO::is_data_available()
- Windows: fix download function with correct path separators
- Avoid creating MLFlow experiment in training algorithm constructor
- ...
v0.9.2
Improvements:
- Update documentation
- Manage error if algorithm code package is not found in Ikomia HUB
- Manage mandatory parameter in IkomiaRegistry::create_algorithm()
- Update tests and make them compatible with PyTest
Bug fixes:
- Fix upper case extension for image and video files
- Fix I/O retrieval for composite I/O like object detection I/O, instance segmentation I/O, ...
- Fix typo error in Workflow::set_parameters()
- ...
v0.9.1
v0.9.0
This release introduces many break changes as we change the naming convention of the Python bindings from the C++ core. The complete API is now in Snake Case style for all functions, there is no more difference between pure Python and C++ API. These changes may occur during the consolidation period of the API marked by the 0.x.x versions.
New features:
- OpenCV 4.7.0
- Python 3.10 support
- MLFlow 1.30
- Remove authentication to install and load algorithms from Ikomia HUB
- Algorithms lazy loading to speed up API initialization
- New base classes for common computer vision tasks:
- object detection
- semantic segmentation
- instance segmentation
- keypoints detection
- Add workflow text I/O
- Add internal algorithms: OpenCV Blur and StackBlur
- Add automatic conversion from instance segmentation I/O to semantic I/O
Improvements:
- New auto-completion system (ik module)
- New documentation
- Algorithm parameters are set from dict structure (not C++ structure anymore)
- Filtering process for object detection, instance segmentation and semantic segmentation outputs
- Handle print() function for many API objects
- System to find best video writer back-end and codec for video export
- displayIO module (now part of ikomia.utils)
- Warn user when an algorithm is not connected to any other algorithm in a workflow
Bug fixes:
- Fix object detection graphics export
- Fix elapsed time computation for training workflows
- Fix JSON serialization of image I/O
- Fix wrong output filenames on batch processing
- Fix synchronous download method for workflow tasks
- ...