On Image to 3D Volume Construction for E-Commerce Applications

Several deep learning techniques have been employed to recover the 3D shape of an object from a single or multiple 2D images. Pix2Vox++ and DeepMesh have been chosen to be investigated as they represent the state of the art techniques for voxel-based and mesh-based techniques respectively. Pix2Vox++ performs poorly on realistic image datasets but well enough on synthetic images. DeepMesh performs well on synthetic datasets but is not yet ready for production when exposed to real-world data. It overfits the synthetic data and is unable to scale well in real-world scenarios.