NeuralDome: A Neural Modeling Pipeline on Multi-View Human-Object Interactions

Figure 1. Our NeuralDome pipeline for processing multi-view video sequences on human object interactions. NeuralDome supports tracking, modeling, and rendering of individual human subjects and objects. To validate NeuralDome, we collect a large dataset HODome over a total of 71 million video frames across 76 viewpoints and process the datasets using NeuralDome for a variety of inference and neural modeling and rendering tasks.