Continuum observations of protoplanetary discs ubiquitously reveal substructures that can be promisingly explained by the presence of embedded forming planets. Characterizing these young systems is key to advancing our understanding of planet formation. Several works have attempted to link the morphology of the observed gaps and rings with the mass of putative planets leveraging numerical simulations that are directly compared with the target observation or used to calibrate empirical formulae. However, these traditional techniques are limited in their computational and time cost or in their ability to account for the full substructure morphology. Additionally, they lack a statistically robust formalization of their estimates which is necessary to properly conduct further data analysis. To overcome these issues, we recently developed and published DBNets: an innovative tool, based on deep learning methods, that can accurately estimate the posterior for the mass of putative planets in observed dust gaps. Nevertheless, the morphology of planet-induced substructures is not only affected by the planet mass but also by other physical properties of the disc. This makes the inference problem strongly degenerate resulting in significant uncertainties on the inferred planet mass. The new release of our tool DBNets2.0, among other improvements, can now deal with this problem by being able to infer the full joint posterior for the planet mass, alpha viscosity, aspect ratio and dust Stokes number. Unveiling these degeneracies provides a way to improve our estimates when prior information on some of the inferred properties is available. In this talk, presenting its unique functionalities and performance, I will show how DBNets2.0 is the way to go whenever you observe dust substructure and wonder about a planet.