Learning Efficient Point Cloud Generation for Dense 3D Object Reconstruction

TLDR

Conventional 3D generative modeling relies on volumetric predictions with 3D convolutions, which are computationally wasteful because 3D shape information is concentrated on surfaces. The authors propose a novel framework that efficiently generates dense point cloud representations of object shapes. The framework uses 2D convolutions to predict 3D structure from multiple viewpoints, jointly applies geometric reasoning with 2D projection optimization, and introduces a differentiable pseudo‑renderer to synthesize depth maps for optimization. Experiments on single‑image 3D reconstruction demonstrate that the method outperforms state‑of‑the‑art baselines in shape similarity and prediction density.

Abstract

Conventional methods of 3D object generative modeling learn volumetric predictions using deep networks with 3D convolutional operations, which are direct analogies to classical 2D ones. However, these methods are computationally wasteful in attempt to predict 3D shapes, where information is rich only on the surfaces. In this paper, we propose a novel 3D generative modeling framework to efficiently generate object shapes in the form of dense point clouds. We use 2D convolutional operations to predict the 3D structure from multiple viewpoints and jointly apply geometric reasoning with 2D projection optimization. We introduce the pseudo-renderer, a differentiable module to approximate the true rendering operation, to synthesize novel depth maps for optimization. Experimental results for single-image 3D object reconstruction tasks show that we outperforms state-of-the-art methods in terms of shape similarity and prediction density.