GP-Adapter: Gaussian Process CLIP-Adapter for Few-Shot Out-of-Distribution Detection
Researchers have developed GP-Adapter, a novel framework designed to enhance the capabilities of CLIP (Contrastive Language-Image Pre-training) models. This method integrates Gaussian Process uncertainty modeling with CLIP's existing architecture to improve performance in few-shot classification and out-of-distribution detection scenarios. By adding class-wise Gaussian Processes on top of frozen CLIP embeddings, GP-Adapter provides variance-aware confidence scores, which are crucial for reliability when dealing with limited data or shifts in data distribution. AI
IMPACT Enhances reliability of vision-language models in low-data and distribution-shifted settings.