BRIXEL applies self-distillation to let vision foundation models like DINOv3 produce higher-quality dense features at fixed input resolution.
SigLIP 2: Multilingual Vision- Language Encoders with Improved Semantic Understanding, Localization, and Dense Features, 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Another BRIXEL in the Wall: Towards Cheaper Dense Features
BRIXEL applies self-distillation to let vision foundation models like DINOv3 produce higher-quality dense features at fixed input resolution.