search

cvlab-kaist / RAIN-GS

Code for "Relaxing Accurate Initialization Constraint for 3D Gaussian Splatting" by Jaewoo Jung, Jisang Han, Honggyu An, Jiwon Kang, Seonghoon Park, and Seungryong Kim
https://ku-cvlab.github.io/RAIN-GS
MIT License
298 stars 16 forks source link

training with sfm and init improvements #12

Closed cv-lab-x closed 6 months ago

cv-lab-x commented 6 months ago

hi, thanks for your great work, have you tested training with sfm points and your init improvements? cause sfm points are very sparse, there's no sfm point on lots of parts of the scene. will the results be more better if using both sfm points and your init improvements ? @crepejung00 Looking forward to your reply, thanks.

crepejung00 commented 6 months ago

Hi, thank you for your interest in our work!

Actually, we didn't try out the setting you questioned and thought it was really interesting! Since most of the datasets that are used for evaluation (e.g., Mip-NeRF360) already contain dense point clouds, we currently do not have the results to answer your question. We will try it as soon as possible and share our results! BTW, we are currently a bit busy due to the Neurips submission deadline and have plans to update our paper and code for another significant performance boost even from random point clouds! Our current timeline is to refactor and release our new code in two weeks, so please stay tuned!

Thanks.

crepejung00 commented 6 months ago

I will temporarily close this issue and reopen it when we are ready with the results to share! Thanks.