The paper introduces the U-Net, a convolutional neural network (CNN) designed for biomedical image segmentation. The authors address the challenge of training deep networks with limited annotated data by leveraging data augmentation techniques, particularly elastic deformations. The U-Net architecture consists of a contracting path to capture context and an expansive path for precise localization, forming a U-shaped structure. This design allows for efficient training with few images and seamless segmentation of large images using an overlap-tile strategy. The network outperforms previous methods in the ISBI challenge for neuronal structure segmentation in electron microscopy stacks and cell tracking in light microscopy images, achieving significant improvements in accuracy and speed. The full implementation and trained networks are available online.The paper introduces the U-Net, a convolutional neural network (CNN) designed for biomedical image segmentation. The authors address the challenge of training deep networks with limited annotated data by leveraging data augmentation techniques, particularly elastic deformations. The U-Net architecture consists of a contracting path to capture context and an expansive path for precise localization, forming a U-shaped structure. This design allows for efficient training with few images and seamless segmentation of large images using an overlap-tile strategy. The network outperforms previous methods in the ISBI challenge for neuronal structure segmentation in electron microscopy stacks and cell tracking in light microscopy images, achieving significant improvements in accuracy and speed. The full implementation and trained networks are available online.