v1.6.0

What's Changed

• Add example python scripts, both local and remote versions. In both cases the attention is visualized.
• Bugfix attention visualization

Full Changelog: v1.5.0...v1.6.0

v1.5.0

What's Changed

• Add custom implementation of calibrator rather than using scikit-learn calibrator. Results are the same but now have no dependency on scikit-learn.
• Add regression test which runs Sybil on random set of NLST images and compares to previously calculated scores

Full Changelog: v1.3.0...v1.5.0

v1.3.0

What's Changed

• Create sybil-predict CLI. So when Sybil is installed, one can use the simple command sybil-predict (Change name from inference -> predict). This also means Sybil can be installed with pipx
• If device not specified, load model onto GPU with the most memory free. Do this when the model is loaded and also at prediction time. Goal here is smarter parallel running. Doesn't affect cpu-only computations.

Full Changelog: v1.2.1...v1.3.0

v1.2.1

What's Changed

• Include pylibjpeg for processing JPEG data within DICOM files
• Update some dependencies to later versions
  Full Changelog: v1.2.0...v1.2.1

v1.2.0

What's Changed

• GH28 Individual inference by @jsilter in #29

Full Changelog: v1.1.0...v1.2.0

Version 1.1.0

What's Changed

• Attention visualization by @pgmikhael in #27
• Download models from GitHub releases instead of Google Drive by @jsilter in #31

Full Changelog: v1.0.3...v1.1.0

Version 1.0.3

What's Changed

• Improve README and install process
• Add a very simple unit test

This release includes example data and checkpoint weights as assets.

Stable Release

Sybil models available for use under MIT license.