We are publishing today a new alpha release of Dafne, after a few iterations to remove bugs and improve usability.
There are some news on the interface side, namely the following:
- Possibility of autosegmenting a batch of slices instead of one-by-one.
- Possibility of erasing mask areas from all ROIs at the same time (useful when regions of the image that are outside the area of interest are incorrectly segmented).
- Command to remove ROI overlap.
- Possibility of initiating a model learning step manually, separately from mask export.
- Possibility of importing a set of masks at the same time (only DICOM format).
Quite a lot of improvements happened "under the hood" and we feel that they improved the overall experience.
The biggest change was the implementation of *blitting*, which is a graphics technique that allows redrawing the areas of the screen that change instead of redrawing the whole image at every frame. This is not hard to implement in theory, but in practice it requires a lot of small details to be taken care of, for example the detection of window resize events. However the implementation seems quite stable now, and it made the interface already much more responsive. If needed, further optimizations are also possible.
The realignment code for the import of masks was also reworked and it is now much faster and overall usable.
**Windows release note**: This windows release was packaged under Windows 10 version 20H2 and might not work correctly with older versions of Windows.