In this paper we discuss two image-based 3D modeling methods based on a multi-resolution evolution of a volumetric function's levelset. In the former the role of the levelset implosion is to fuse ("sew" and "stitch") together several partial reconstructions (depth maps) into a closed model. In the latter the levelset's implosion is steered directly by the texture mismatch between views. Both solutions share the characteristic of operating in an adaptive multiresolution fashion, in order to boost up computational efficiency and robustness.