The nanomechanics of crystalline materials is governed by various processes occurring across a wide range of length/time scales. While continuum mechanics simulations offer a suitable framework for modeling the mechanical properties of microstructures, they are limited in capturing the details of crystal defect structures and their elementary deformation mechanisms. On the other side, these mechanisms can be conveniently accessed through discrete atomistic simulations, but such approaches are usually limited to small length/time scales. Now more than ever, accurate information transfer across scales is required for efficient and reliable physics-based nanomechanics modelling. The work we present here explores a novel atomistic-to-continuum crossover scheme based on dislocation density fields. More precisely, elastic transformation tensors are computed using the Hartley and Mishin method for atomistic configurations, and then employed as inputs in a micromechanical field dislocation mechanics (FDM) strain-gradient type model using a regular fast Fourier transform solver grid. This versatile approach successfully captured, defects as diverse as dislocations and high-angle grain boundaries, as well as interactions among them, in both cubic and hexagonal crystals. Assessments of this approach with cubic (Cu, Al) and hexagonal (Mg, Ti) materials is also presented. The prediction by means of machine learning approaches of interfaces characteristics as represented by dislocation density fields is also explored. In the light of our results, the implication of such a discrete-to-continuum crossover for bridging scales in nanomechanics is discussed.