Four training geometric models as imaged by each imaging method. The first row shows the visibility amplitudes of the model (orange) compared to the visibility amplitudes measured for M87^* (blue). The second row shows the ground-truth images. The DIFMAP, eht-imaging, and SMILI rows show a fiducial image made from the same parameter sets as the images shown in Fig. 17. The THEMIS and Comrade rows show a random draw from the posterior. The circle in the DIFMAP panels represent a Gaussian blurring kernel of 20 μas. The white lines in the THEMIS and Comrade panels represent the size of FWHM for the Gaussian blurring kernel needed to match the DIFMAP effective resolution.