Synthesizing a Large Scene with Multiple NeRFs.

Last several years, NeRF achieved great success in view synthesis since it can render high-quality images in a complex scene. However, we find that its ability to rebuild a large scene is low because images that are far apart and do not overlap with each other will affect each other in the training process. In order to solve this problem, we propose Cluster-based NeRF, which splits the original input images into several clusters and then train a NeRF for each cluster. We also design an algorithm to improve the rendering quality in the overlapping areas. In the experiments, we show that our method outperforms the traditional NeRF on both the blender and real world dataset.