#test-time adaptation

Batch-Conditioned Semantic Anchors Enable Robust TransductiveNEW⭐9

JPTA is a prior-aware framework for robust transductive adaptation of vision–language models via batch-conditioned semantic reference estimation.