Z2P: Instant Visualization of Point Clouds

We present a technique for visualizing point clouds using neural network. Our allows an instant preview of any cloud, and bypasses the notoriously difficult surface reconstruction problem or need to estimate oriented normals splat-based rendering. cast as conditional image-to-image translation task, design network that translates depth-map directly into image, where cloud is visualized though was reconstructed from it. Furthermore, resulting appearance can be, optionally, conditioned on simple control variables (e.g., color light). demonstrate our instantly produces plausible images, can, on-the-fly effectively handle noise, non-uniform sampling, thin surfaces sheets.