Elaboration on Occupancy Networks and Resizing Strategy

Occupancy networks introduces the decision boundary of neural network classifier to express arbitrarily complex three-dimensional topological structure, thereby solving the problem of incompatibility between the efficiency of calculation and the validity of storage in the field of 3D reconstruction. We briefly introduced the principle of occupancy network and its single-object 3D reconstruction effect. When further applying occupancy network to real online images, we propose a feasible resizing strategy to improve the versatility of occupancy network. We believe this will help occupancy network move towards a broader scene.