Fig. 1: Schematic drawing shows the layout of RTU-Net.

a Simplified schematic of RTU-Net training procedure. The high-resolution 3D stationary samples can be trained using synthetic or experimental methods. The network is optimized with three loss functions. b The schematic shows the prediction procedure of RTU-Net. c Examples highlights the working scale of RTU-Net, including structural imaging of mitochondria and microtubule proteins in microscale (left), neuroimaging of the rat brain in mesoscale (center), and PIV measurement for particle imaging in macroscale (right)