Our novel method improves visual prompting accuracy through affine/color transformations and TrivialAugment data augmentation, achieving state-of-the-art results with minimal overhead.

We propose a probabilistic model using Gaussian processes for learning regression functions from data without correspondence.