This repository provides the PyTorch implementation of Deformable Attention Transformer (DAT), a CVPR 2022 paper achieving state-of-the-art results on image classification benchmarks. The method applies learned offset-based deformable attention to Vision Transformers, enabling dynamic receptive fields that adapt to spatial content while maintaining linear computational complexity. The codebase includes training and evaluation code for image classification, with related repositories for object detection and segmentation tasks.