rogermm14 / eonerf

Earth Observation NeRF
https://rogermm14.github.io/eonerf
24 stars 1 forks source link

question/request for masks #5

Open ryansvf opened 5 months ago

ryansvf commented 5 months ago

Hello,

Thanks for presenting this awesome work! I had a few questions on part of the paper.

In Section 4.2, you stated: "It should be noted that the MVS surface models in Figures 8 and 9 present incomplete regions (in black) beyond the water zones (also in black), which we ignore to compute the altitude MAE. As explained in Figure 9, for a fair comparison, we remove surface points non-observed by the majority of cameras before computing the altitude error."

and in the Figure 9 description: "...(b) EO-NeRF prediction at points observed by more than 3/4 of all input cameras...The altitude MAE is computed using the surface points in (b)."

My questions from this are:

  1. Are you creating a mask from the incomplete regions and applying it to your methods (EO-NeRF and Sat-NeRF)? or just to PSM? From the section 4.2 statement mentioned above, it seems like its being applied only to PSM, but the Fig 9 description suggests that it's being applied to EO-NeRF predictions as well.
  2. I understand that you are masking out water regions from the CLS tiffs provided through IEEE DFC19, but how are you computing these secondary masks based on incomplete regions?
  3. Can you make these "incomplete region" masks available on your google drive so that I and others can use them for our evaluation?
  4. Can you make the final EO-NeRF DSMs (no water masks nor "incomplete region" masks applied) available on your google drive as well?

Thanks, Ryan

ryansvf commented 2 months ago

pinging this again to see if we can get these masks?