Bin-wise mean redshift residuals and σ_MAD (defined in the main text) for all the image translation models. For each translation model, the redshift predictions were made for the generated test images using redshift estimation models trained with the generated ɀ-training images. A comparison is made with the predictions for the original SDSS test images by a model trained with the original SDSS ɀ-training images (i.e. the baseline). First row: Bin-wise mean redshift residuals as a function of spectroscopic redshift ɀ_spec for the S2S and S2C translations, separately. The error bars are estimated using the root mean square error of residuals in each redshift bin. Second row: Bin-wise mean redshift residuals as a function of the predicted redshift point estimate ɀ_est. Third row: Bin-wise mean redshift residuals as a function of r-band Petrosian magnitude. Fourth row: MAD-based dispersion, σ_MAD, as a function of r-band Petrosian magnitude. The curve obtained by the Swin Transformer in the S2S translation almost overlaps with the baseline.