search

Loping151 / ForPlane

Neural LerPlane Representations for Fast 4D Reconstruction of Deformable Tissues
Other
42 stars 2 forks source link

Fail to exatract meshes given a trained model #8

Closed ysjue closed 1 day ago

ysjue commented 7 months ago

Hi, appreiciate this fantastic work! After training the models, I tried to extract a mesh from the predicted density fields (kind of like what EndoNeRF and EndoSURF did in their paper) using Marching Cubes. However, I noticed that the predicted density field has no 0-density (0-sigma) grid, which disables Marching Cubes to extract surfaces effectively. No matter how I tuned the threshold, I couldn't extract any meshes visually making sense. May I ask if these observations are normally expected and not caused by some bugs in my implementation? Does this issue happen because the authors sample the rays in an efficient manner so that density predictions in unoccupied space are insufficiently penalized? Thanks!

Loping151 commented 7 months ago

We haven't tested extracting mesh yet, but we believe the density distribution is accurate because the Forplane is able to provide sufficiently accurate depth maps.