When 3 D Reconstruction

3D reconstruction from a single image is a classical problem in computer vision. However, it still poses great challenges for single-view reconstruction of daily-use objects with irregular1 shapes. In this paper, we re-consider the problem of single-view 3D reconstruction in terms of two CV areas: category modeling and knowledge mining from big visual data. Category modeling & task output: For the bottleneck in single-view 3D reconstruction, i.e. the reconstruction of objects with irregular1 structures, we have to return to the concept of “category modeling”. Therefore, the objective is to train a model to detect objects in the target category in large images, while simultaneously projecting their 2D shapes into the 3D space at the pixel level. The category model encodes the knowledge of how intra-category structure deformation and object rotations affect 2D object shapes. Mining from big visual data & task input: Another bottleneck lies in efficiently learning the category-specific knowledge of 3D reconstruction for a huge number of categories. Ideally, we would need to train a model for each object category in daily use, so as to construct a knowledge base to provide a 3D reconstruction service for arbitrary RGB images. Therefore, we hope to learn from big visual data to avoid the labor of manually preparing training samples (e.g. well built 3D models) for each category, and thus ensure a high learning efficiency.