C-Flow: Conditional Generative Flow Models for Images and 3D Point Clouds

We propose C-Flow, a conditioning scheme for flow-based generative models applicable to many different domains. The figure shows the results of modeling the conditional distributions image ↔ 3D point cloud. In the top row we apply this model for 3D reconstruction (image → point cloud), and in the bottom row for rendering new images (point cloud → image). Our model allows sampling multiple times from this conditional distribution to generate several renderings of the same point cloud.