PhD student, École de technologie supérieure, Université du Québec
1 paper at NeurIPS 2025
We propose CLIPTTA, a contrastive test-time adaptation method for CLIP that improves both accuracy and OOD detection in closed- and open-set settings.