It has been a long time since the last Dafne update, but we are back with a big one!
First of all, this is not really new, but check out the Dafne paper published in Radiology AI. Please cite this paper in your future works!
Not much has changed aesthetically, but Dafne is very different under the hood, and it’s ready to tackle new challenges. We have now support for 3D models, instead of the original Dafne 2D models, and they can also be based on PyTorch, instead of Tensorflow. Actually, I predict that PyTorch will be the way to go in the future. But we are keeping the possibility for both.
The new models can also depend on external packages hosted on PyPI. These packages are installed at runtime, giving us a fantastic degree of extra freedom moving forward.
For this reason, Dafne is now installed differently than before. There are new installers for Mac and Windows, but they are actually scripts that install Python, and then Dafne as a Python script. They are available from the Github releases page. You will need to uninstall the previous version of Dafne before installing the new one! You will not lose your settings and data.
This also means that Dafne can auto-update itself, so you won’t need to download a new installer every time.
Another piece of news is that we are moving towards a new server architecture. This new architecture allows us bigger models, more flexibility, and finer access control. Not much should change from the user’s side, but please make sure to upgrade Dafne, as the new server is not compatible with older Dafne versions. Your API key should remain valid on the new server. We are planning to do the switch towards the end of the month.
Finally, we are happy to announce that Aschoplex for the segmentation of the Choroid Plexus is now part of Dafne and will be offered on the new server. This is thanks to Valentina Visani, who is the original developer.
More news are coming soon, so upgrade now and stay tuned!