search

maturk / dn-splatter

DN-Splatter: Depth and Normal Priors for Gaussian Splatting and Meshing
https://maturk.github.io/dn-splatter/
Apache License 2.0
404 stars 25 forks source link

Plans to Move to Nerfstudio #63

Open peasant98 opened 2 weeks ago

peasant98 commented 2 weeks ago

Are there plans to move this repo to nerstudio? It is nice that it is built on top of nerfstudio, but it is not in the main repo itself and might be harder for others to find. I think there are some nice quality of life add ons that would really improve splatfacto in the main repo.

maturk commented 2 weeks ago

@peasant98 thanks for the interest. I think at the very moment, this repo is more of a research playground and I do not intend to make it a part of Nerfstudio oficially mainly because maintenence reasons. I also believe most users are interested in monocular depth supervision and unfortunately the methods implemented in this repo currently work the best with real depth sensors. I do think this repo can be useful for researchers, but perhaps not the more casual users of Nerfstudio.

What do you think @peasant98? Did you have some ideas or perhaps suggestions for me to look at? I will update this repo shortly with improvements from another research project we have worked on, but again, this was targeting sensor depths and not the more general (but difficult) mono depth variant.

peasant98 commented 2 weeks ago

Hi @maturk , I think this would be really useful for the folks of nerfstudio to use. The integration of depth and normals help quite a bit including in the few shot case (which I have seen myself), and are a nice quality of life thing that can help when someone has SfM points and real sensor data. We can even supervise on depths with losses like Pearson correlation coefficient that don't require a metrically grounded depth.

After a conference deadline, I would be super down to integrate some of the stuff here into Nerfstudio!

maturk commented 6 days ago

Hey @peasant98, just to keep you updated, we have a few updates in #64 and we also plan to add some more features probably within the next month or so.