Fiducio: paper accepted at TMLR, library released

Our paper Rethinking Post-Hoc Calibration in Semantic Segmentation has been accepted at Transactions on Machine Learning Research (TMLR). It studies two principles for post-hoc calibrators: translation invariance and decision preservation.

Alongside the paper we release Fiducio, an open-source Python library implementing the calibrators studied there, for classification as well as 2D and 3D semantic segmentation.

• 📄 Paper & project page: fiducio-ai.github.io/Fiducio/paper
• 💻 Code: github.com/fiducio-ai/Fiducio
• 📦 Install: pip install fiducio
• 🎉 Announcement on LinkedIn

Joint work with Kim-Celine Kahl, Bálint Kovács, Maximilian Rokuss, Klaus Maier-Hein, Xavier Coubez, Philippe Meyer and Sylvain Faisan — ICube Strasbourg, Université de Strasbourg, Institut Strauss and MIC @ DKFZ.