Generating three-dimensional models from two-dimensional free-hand sketches presents a viable avenue for 3D reconstruction. As a convenient method for 3D reconstruction, sketch-based modeling not only significantly eases the challenges associated with 3D model generation, but also affords designers the flexibility to craft models in alignment with their own conceptual visions. Nevertheless, the intrinsic abstraction and diversity of free-hand sketches, which often lack the rigorous geometric and physical constraints typical of actual images, exacerbate the difficulty in adapting models to various sketch styles. This has resulted in an underexplored methodology within the field. In this study, we introduce PotC2Vox—a novel neural network architecture—whereby a two-dimensional point cloud, derived from preprocessed sketches, is supplied to the network. The network adeptly captures geometric features through dedicated processing modules, utilizing the 2D point cloud to extract image characteristics and to execute the 3D reconstruction. Furthermore, we integrate a Contextual Modal Enhancement Module (CME) into our network, establishing a link between the input features at the expense of minimal network depth, thereby further refining the 3D reconstruction outcomes. Comparative evaluations with a suite of experiments demonstrate that our PotC2Vox model exhibits superior performance in the reconstruction quality and fidelity of 3D models relative to analogous networks.