Cellular material, namely foams or lattices, are characterized by complex architecture where the material is concentrated in small spans or thin webs, at an intermediate scale between the constituents and the structure. This particularity provides these materials with remarkable specific properties. The recent development of new manufacturing processes (especially additive manufacturing), pushes these materials to a new stage. Many open questions are stirring up the scientific community, in particular, multiscale modelling, characterization of the associated parameters, fast and accurate simulation, and geometric hazard. In essence, foams have a random architecture, thus their geometry is locally variable and unknown. For lattices, defects inherent to additive manufacturing processes are known to introduce non-negligible geometric biases. In view of identifying the mechanical properties, the geometric characterization of the sample is therefore of paramount importance to build a digital mechanical twin associated to each individual tested sample. In this talk, we present different types of image-based modelling techniques and how do they help to bridge mechanical experiments and numerical simulations. We will then focus on techniques that build an explicit geometry based on splines and the corresponding control point fitting procedure. In order to achieve the lightest possible model, and given the morphology of the materials considered, the method aims at building explicit isogeometric beams and isogeometric volumetric models. These models are called analysis-suitable in the sense that they are ready for mechanical simulations. To do so, we propose to extend the Virtual Image Correlation algorithm to the case of branching beams and branching surfaces. We also aim at characterizing the potentially evolving cross-section thickness of each beam. Note finally that we also propose automatic initialization strategies in situation where the topology is unknown a priori, such as foams. Technically, the sample is described as a set of interconnected quadratic B-Spline beams or surfaces. Two (2D) or three (3D) degrees of freedom (dof) are associated with each control point position. Additional degrees of freedom may be introduced to describe the thickness of the beams. All these dofs are adjusted to minimize a distance between the real image and a synthetic virtual image computed as a function of the dofs.