UnInfo adapts CLIP at test time by updating a low-rank adapter on the image encoder with a combination of confidence maximization, uniformity loss and EMA distillation, improving zero-shot accuracy under image corruptions.
Improving zero- shot generalization for clip with synthesized prompts,
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
-
Uniformity First: Uniformity-aware Test-time Adaptation of Vision-language Models against Image Corruption
UnInfo adapts CLIP at test time by updating a low-rank adapter on the image encoder with a combination of confidence maximization, uniformity loss and EMA distillation, improving zero-shot accuracy under image corruptions.