We propose a novel method to reconstruct and render dynamic scenes including dense scene flow, establishing correspondences over time.
We propose the currently best performing method to reconstruct object faithfully for all possible lighting environments.
We present a new radiance field representation that can be used to create photorealistic assets from a simple smartphone video that can be rendered volumetrically in real-time in state-of-the-art game engines. Rendering is comparably fast with mesh rendering, but includes volumetric effects and is suitable for fur, hair, ...
We present a new radiance field representation that can be semantically decomposed into different classes, with applications in 3D segmentation and editing.
We propose a strategy to assess overfitting using gradient and weight statistics in neural networks. This makes the use of a separate validation set unnecessary.