Releases: smistad/FAST
v3.0.0
Several major changes in this new release, including some API changes, therefore this marks a new major release to version 3.
The major changes include:
- Refactoring of the pipeline core, DataPort is now called DataChannel
- Support for more inference engines using a dynamic linking system (OpenVINO, TensorRT, TensorFlow cuDNN, ROCM and CPU)
- Patch generator and patch stitcher for patchwise processing of large images
- Refactored neural network code. Renamed PixelClassifier to SegmentationNetwork and ImageClassifier to ImageClassificationNetwork.
- ImageToBatchGenerator to turn a stream of images/patches to a stream of batches of a given size.
- Frame data system which are metadata that are copied from input to output data of a process object.
- Support for whole slide microscopy images: FAST uses OpenSlide to open these images and a custom ImagePyramidRenderer to visualize the images efficiently.
- More test data (WSI, DICOM CT thorax, more neural network models)
- 3 neural network examples
- FAST releases now include the Clarius listen API, Intel OpenVINO and RealSense
- SegmentationLabelRenderer
Release binaries can be downloaded below, by pressing "assets". Please make sure you have the required dependencies installed on your system before using the release.
v3.0.0-rc4
Several major changes in this new release, including some API changes, therefore this marks a new major release to version 3.
The major changes include:
- Refactoring of the pipeline core, DataPort is now called DataChannel
- Support for more inference engines using a dynamic linking system (OpenVINO, TensorRT, TensorFlow cuDNN, ROCM and CPU)
- Patch generator and patch stitcher for patchwise processing of large images
- Refactored neural network code. Renamed PixelClassifier to SegmentationNetwork and ImageClassifier to ImageClassificationNetwork.
- ImageToBatchGenerator to turn a stream of images/patches to a stream of batches of a given size.
- Frame data system which are metadata that are copied from input to output data of a process object.
- Support for whole slide microscopy images: FAST uses OpenSlide to open these images and a custom ImagePyramidRenderer to visualize the images efficiently.
- More test data (WSI, DICOM CT thorax, more neural network models)
- 3 neural network examples
- FAST releases now include the Clarius listen API, Intel OpenVINO and RealSense
Release binaries can be downloaded below, by pressing "assets". Please make sure you have the required dependencies installed on your system before using the release.
v3.0.0-rc3
Several major changes in this new release, including some API changes, therefore this marks a new major release to version 3.
The major changes include:
- Refactoring of the pipeline core, DataPort is now called DataChannel
- Support for more inference engines using a dynamic linking system (OpenVINO, TensorRT, TensorFlow cuDNN, ROCM and CPU)
- Patch generator and patch stitcher for patchwise processing of large images
- Refactored neural network code. Renamed PixelClassifier to SegmentationNetwork and ImageClassifier to ImageClassificationNetwork.
- ImageToBatchGenerator to turn a stream of images/patches to a stream of batches of a given size.
- Frame data system which are metadata that are copied from input to output data of a process object.
- Support for whole slide microscopy images: FAST uses OpenSlide to open these images and a custom ImagePyramidRenderer to visualize the images efficiently.
- More test data (WSI, DICOM CT thorax, more neural network models)
- 3 neural network examples
Release binaries can be downloaded below, by pressing "assets". Please make sure you have the required dependencies installed on your system before using the release.
v3.0.0-rc2
Several major changes in this new release, including some API changes, therefore this marks a new major release to version 3.
The major changes include:
- Refactoring of the pipeline core, DataPort is now called DataChannel
- Support for more inference engines using a dynamic linking system (OpenVINO, TensorRT, TensorFlow cuDNN, ROCM and CPU)
- Patch generator and patch stitcher for patchwise processing of large images
- Refactored neural network code. Renamed PixelClassifier to SegmentationNetwork and ImageClassifier to ImageClassificationNetwork.
- ImageToBatchGenerator to turn a stream of images/patches to a stream of batches of a given size.
- Frame data system which are metadata that are copied from input to output data of a process object.
- Support for whole slide microscopy images: FAST uses OpenSlide to open these images and a custom ImagePyramidRenderer to visualize the images efficiently.
- More test data (WSI, DICOM CT thorax, more neural network models)
Release binaries can be downloaded below, by pressing "assets". Please make sure you have the required dependencies installed on your system before using the release.
v3.0.0-rc1
Merge branch 'master' of github.com:smistad/FAST
v2.1.0-rc1
Major changes
- Supported for RealSense streaming
v2.0.0
Only downloads marked "with-neural-network" has the neural network module with tensorflow enabled. Downloads with cudnn in the name, enables GPU accelerated tensorflow, this requires CUDA 9 and cuDNN 7 installed.
Due to the extent of this update and several breaking changes to the API, the major version of FAST is bumped up to 2.
Major changes:
- Improved execution pipeline with focus on streaming and synchronization
- Updated all renderers to use modern OpenGL (3.3)
- Major refactoring of the Mesh data objects, importers, exporters and renderers
- Rewrote 2D rendering completly
Minor changes/additions:
- Added MovieStreamer, which can stream frames from movies using Qt5 multimedia
- HounsefieldConverter PO
- Updated tensorflow to version 1.8
- UltrasoundImageEnhancement PO
- Dicom image importer using DCMTK
- Updated Qt to version 5.9.4
- +++
Known issues:
- Pipelines are not always stopped properly, resulting in a deadlock.
- FAST is unstable on Mac, this seems to be due to Apple's buggy OpenCL implementation.
v1.2.0
Changelog
- Fixed cmake rpaths enabling Mac binary release
- Renamed PixelClassification to PixelClassifier, PointRenderer to VertexRenderer and MeshRenderer to TriangleRenderer
- Fix lung segmentation example on windows
- Windows shared library (DLL)
- SimpleGUI example fixed
- qt interop example fixed
- Non local means example fixed
- Doxygen api documentation
- Scaling of fonts on large screens
- Pipeline editor
- ++
v1.1.0
v1.0.0
This is the first release of FAST with binaries for linux and windows.
Major changes from previous release includes:
- TensorFlow integration, enabling neural network pipelines running image classification and segmentation.
- Reduction in number of dependencies (boost and glew has been removed)
- Text pipelines allow users to create and modify pipelines without recompiling FAST
- OpenIGTLinkServer and OpenIGTLinkClient tools
- Kinect support using libfreenect2 (this module is not enabled in the binary release)
- Several new algorithms
- +++