Abstract: Highlights•A self-Attention and Convolution hybrid module is proposed for planar reconstruction.•Explicit connection between adjacent layers is built for efficient information flow.•A line segment filtering method is presented to refine planar guiding information.
Loading