Recent advances in 3d deep learning have shown that it is possible to train
highly effective deep models for 3D shape generation, directly from 2D images.
This is particularly interesting since the availability of 3D models is still
limited compared to the massive amount of accessible
论文提出了一种统一框架,用于解决单幅图像特定类别的 3D 重建和新 3D 形状生成的问题。该方法支持弱监督学习,只需要单个实例的 2D 图像,使用网格作为输出表示,并利用光照信息提高了性能。实验结果表明,该方法在定量度量上与最新的基于体素方法相当或优于,而且结果更加美观,并且在弱监督学习方面表现良好。