RegNeRF: Regularizing Neural Radiance Fields for View Synthesis from Sparse Inputs
@inproceedings{niemeyer2022regnerf,
title={Regnerf: Regularizing neural radiance fields for view synthesis from sparse inputs},
author={Niemeyer, Michael and Barron, Jonathan T and Mildenhall, Ben and Sajjadi, Mehdi SM and Geiger, Andreas and Radwan, Noha},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={5480--5490},
year={2022}
}
Improvement over NeRF in the few-shot context (only few input images). It does better than NeRF in that context because it focusses on a consistent 3d representation. They do this by including a Loss for geometric/depth smootheness.
