This repository provides the official PyTorch implementation of CrossViT, a vision transformer architecture that combines multi-scale features through cross-attention mechanisms for improved image classification on ImageNet. The implementation includes training scripts, pretrained model weights, and supports distributed multi-GPU training.