Cross-pattern in the fitting results

[mkuemmel]

I ran SE++ with sersic fitting on simulated images with sersic profile (in the context of the Euclid Morphology Challenge). When plotting the difference between two SE++ parameters pixel_centroid_x/y and the x/y, which are the fitted object positions, the following pattern emerges:

The boxy from of the difference an be explained by the sub-pixel interpolation scheme, the cross-like pattern remains a mystery. Some more information:

• detection and the only measurement image are identical, hence both quantities are determined on the same image;
• the image PSF is fairly symmetrical and does not contain any cross-like pattern such as diffraction spikes or similar
• in other scenarios such as fitting a disk+bulge or disk+sersic model to disk+bulge simulations shows the boxy structure, but not this pronounced cross.