In recent years, transformer-based architectures have revolutionized computer vision, achieving state-of-the-art results across diverse tasks. Nevertheless, conventional Vision Transformers (ViTs) demand massive datasets and substantial computational resources, which restricts their applicability in resource-constrained or data-scarce scenarios. To overcome this limitation, DataEfficient Image Transformers (DEiTs) have been introduced, leveraging three key strategies: (i) knowledge distillation, where a compact student transformer benefits from the guidance of a powerful CNN teacher; (ii) optimized training protocols, including regularization and augmentation tailored for small data regimes; and (iii) efficient architectural modifications that reduce redundancy while preserving representational power. By integrating these mechanisms, DEiTs not only narrow the gap between transformers and CNNs under limited data conditions but also establish a scalable framework for future vision applications requiring both efficiency and accuracy.