We present an architecture for rendering multiple views efficiently on a cluster of GPUs. The original scene is sampled by virtual cameras which are used later to reconstruct the desired views. We show that this image-based approach can be very scalable and support rendering at interactive rates.
展开▼
