Depth Estimation for epipolar transformation

dcharatan / pixelsplat

[CVPR 2024 Oral, Best Paper Runner-Up] Code for "pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction" by David Charatan, Sizhe Lester Li, Andrea Tagliasacchi, and Vincent Sitzmann

http://davidcharatan.com/pixelsplat/

MIT License

830 stars 56 forks source link

Depth Estimation for epipolar transformation #77

Closed Eyvaz27 closed 3 months ago

Eyvaz27 commented 3 months ago

Hello, I have observed that when we are increasing the resolution of the images and count of images, time requirements increase almost exponentially due to intersect_rays function in src.geometry.projection.py. Would it be useful to use something like https://math.stackexchange.com/questions/1993953/closest-points-between-two-lines/4764188#4764188 to get direct depth estimates

dcharatan commented 3 months ago

Have you profiled the code to make sure that intersect_rays is actually what's causing a substantial slowdown? If you find that the approach from Stack Exchange is much faster, I would be happy to accept a pull request!