Adapting CLIP Models Without Training: Exploring a Gaussian Discriminant Analysis (GDA) approach for downstream classification of CLIP, this research offers a training-free method enhancing CLIP’s performance in various tasks. Discover more at arXiv:2402.04087.

• Leverages knowledge from visual and textual modalities.
• Surpasses state-of-the-art results in certain scenarios.

The exploration into GDA provides an efficient way to utilize pre-trained models like CLIP, especially in situations where additional computational resources are not feasible.