Efficiently train high-resolution vision transformers using a multi-window strategy for improved performance and reduced training costs.
