Developed at the ENS Paris-Saclay, Centre Borelli and accepted at the CVPR EarthVision Workshop 2023.
Roger Marí, Gabriele Facciolo, Thibaud Ehret
Abstract: We introduce EO-NeRF, the Earth Observation NeRF. This method can be used for digital surface modeling and novel view synthesis using collections of multi-date remote sensing images. In contrast to previous variants of NeRF for satellite imagery, EO-NeRF outperforms the altitude accuracy of advanced pipelines for 3D reconstruction from multiple satellite images, including classic and learned stereo-based methods. This is largely due to a rendering of building shadows strictly consistent with the geometry of the scene and independent from other transient phenomena. A number of strategies are additionally presented with the aim to exploit satellite imagery out of the box, without requiring usual pre-processing steps such as a relative radiometric normalization of the input images or a bundle adjustment of the associated camera models. We evaluate our method on different areas of interest using sets of 10-20 true color and pansharpened WorldView-3 images.
If you find this code or work helpful, please cite:
@InProceedings{Mari_2023_CVPR,
author = {Mar{\'\i}, Roger and Facciolo, Gabriele and Ehret, Thibaud},
title = {Multi-Date Earth Observation NeRF: The Detail Is in the Shadows},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2023},
pages = {2034-2044}
}