Focal Stack

Focal stacking is a computer vision technique that reconstructs a fully focused image from multiple images captured at varying focal distances (a "focal stack"). Current research focuses on improving depth estimation from these stacks using deep learning models, including convolutional neural networks (CNNs) and transformer architectures often incorporating LSTM modules for handling variable-length stacks. These advancements aim to overcome limitations of traditional methods, such as sensitivity to noise and reliance on specific camera settings, leading to more robust and accurate depth maps for applications in photography, robotics (e.g., autonomous racing), and light field imaging. The development of larger, higher-resolution datasets and self-supervised learning techniques are also key areas of progress.