Abstract: Most existing traditional and deep learning (DL)-based methods used for random noise attenuation or reconstruction of seismic data typically only process 2-D data. Very few methods are able ...