OpenOOD
OpenOOD copied to clipboard
add SAL (Separate And Learn) method
There's this new paper: How Does Unlabeled Data Provably Help Out-of-Distribution Detection? by Xuefeng Du, Zhen Fang, Ilias Diakonikolas, and Yixuan Li Performance is quite good according to their paper and they have published their source here https://github.com/deeplearning-wisc/sal Would be interesting how it compares to other methods.
Seems a promising work. Would you be interested to adapt it to OpenOOD?
Unfortunately I'm currently too busy, maybe in a few months.