search

centreborelli / satnerf

Satellite Neural Radiance Fields
https://centreborelli.github.io/satnerf/
BSD 3-Clause "New" or "Revised" License
127 stars 25 forks source link

Poor DSM within water areas #5

Closed wuy202 closed 2 years ago

wuy202 commented 2 years ago

Quality of DSMs generated using Sat-NeRF is pretty bad when it comes to water areas, any solution to such problem and what might be the reason that cause this error ?

rogermm14 commented 2 years ago

Hello, As explained in the papers, NeRF and Sat-NeRF require radiometric consistency to infer geometry. Water areas have no radiometric consistency, so it should not be a surprise that the geometry is erroneous there. If you are using the DFC2019 data, semantic segmentation is available including water labels. If you are using your own data, I would suggest that you first detect the water areas in your satellite images and then mask the output NeRF/SatNeRF DSMs accordingly.