What is Vision Transformer (ViT) and how does it process images?

ViT applies Transformer architecture to image patches instead of convolutional features.

Vision Transformer divides an image into fixed-size patches, flattens them, and projects them into embeddings. These embeddings are treated as tokens and passed through Transformer encoder layers using self-attention. Unlike CNNs, ViT relies entirely on global attention, enabling better long-range dependency modeling but requiring large datasets for training.