#machine learning
Jul 2026
Rethinking Post-Hoc Calibration in Semantic Segmentation
This work characterize translation-invariant (TI) calibrators and construct TI counterparts of shift-sensitive methods and introduces more expressive class-conditional affine calibrators that preserve decisions under argmax- and order-preservation constraints.
Tristan Kirscher, Kim-Celine Kahl, Balint Kovacs et al.
· arXiv.org · 0 citations