|
Figure 2
Workflow for creating a data pair to train pix2pix-cGANs. 3D cubes with and without air artifacts are shown in (a) and (b), respectively. Using a dedicated image processing pipeline, air artifacts are identified and extracted. The extracted air artifacts, illustrated in (c), are copied to a cube without air artifacts (b). The resulting 3D cube with copied air artifacts is demonstrated in (d). This workflow creates source (d) and target (b) pairs for training the pix2pix-cGANs. The length of each cube is 256 pixels in all directions. |