SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features

Tschannen, Michael, Gritsenko, Alexey, Wang, Xiao · arXiv · 2025