Sketch of the main part of the SPECTROTRANSLATOR algorithm. The difference between the spectroscopic parameters of catalogue A and B (∆X_a→b) are computed in a deep fully connected network, where the inputs are composed of the spectroscopic parameters from catalogue A, and of photometric colours. The deep network of depth-n is made by a series of n blocks of progressively increasing complexity, until a central layer of maximum complexity, before decreasing symmetrically. Each block is composed of a REsNET-like unit (see Sect. 2.1), illustrated on the right part, and of a linear layer that adjust the number of feature (nf ) between each depth. Finally, the difference predicted by the deep network are added to the spectroscopic parameters of catalogue A to compute the parameters values expressed in the base of catalogue B.