Photometric 3D-reconstruction techniques aim at inferring the 3D geometry of a scene, given one or several 2D images of this scene. They do so by analyzing the luminous quantities in the images. A physics-based model for the observed colors is introduced, which describes the interactions between the geometry, the reflectance of the observed surface, the lighting and the camera. Then, this model needs to be inverted in each pixel so as to recover the geometry. In this talk, we describe some recent advances regarding modeling and resolution of shape-from-shading (single image) and photometric stereo (multiple images obtained under varying lighting). We show that resorting to PDEs is a natural way to cope with some of the arising ambiguities, but also to elegantly formulate the problem by reducing the number of unknowns and priors. Nevertheless, the resulting PDEs are usually nonlinear and cannot be solved exactly because of measurement noise. Hence, we discuss possible linearization strategies to make the problem tractable, as well as variational resolution schemes for ensuring robustness. Works in collaboration with Jean-Denis Durou (Université de Toulouse, France).