Fig. 4.

The pipeline for the training and iterative reconstruction processes. During the training process, $P{(\Lambda )}_1$ is employed to extract the limited-sampled information of 240° indicated by the white part, while disregarding the information of the black. The reconstruction process comprises two reverse SDEs, involving the predictor-corrector (PC) and data consistency (DC). Similarly, $P{(\Lambda )}_1$ is utilized to extract the limited-sampled information of 240° indicated by the white part, while ignoring the information of 120° indicated by the black part. Conversely, $P{(\Lambda )}_2$ is used to extract the limited-sampled information of 120° indicated by the white part, disregarding the information of 240° indicated by the black part.