Results from our transfer learning method. When we train models on SED Wide data, and use them to select quiescent galaxies from Int Wide data (dashed orange curve), the F1-scores are only slightly lower than those we obtain from models trained on Int Wide data (solid blue curve). A generally similar result is obtained when we train models on Int Wide data and use them to select quiescent galaxies from SED Wide data (dotted black curve). For comparison we also show the results from the LePhare template fitting method applied to the SED Wide mock catalogue (see Sect. 6.4), using Euclid, ugriz, W1, W2, and 20 cm photometry, and with redshifts fixed at the Laigle et al. (2016) values.