Video wird geladen...
Video konnte nicht geladen werden
Big day for open source 3D capture! 🌐 nerfstudio hits 1.0 and gets a *commercial-ready* Gaussian Splatting implementation Google AI open sourced Zip-NeRF — the highest quality 3D reconstruction pipeline, setting the stage for a future SMERF release
85,222 Aufrufe • vor 2 Jahren •via X (Twitter)
10 Kommentare

More understated, but equally exciting is Google’s CamP — which dramatically improves 3D pose estimation of your input 2D images, giving you much cleaner 3D reconstructions than using COLMAP. Code linked below (oh and it’s an Apache license too!) ⬇️

Zip-NeRF is SOTA for visual quality but not speed. Perhaps we‘ll see that SMERF release later this year bringing us quality AND speed? Needless to say, 2024 is shaping up to be a fun year for 3D reconstruction :) More in my video below ⬇️

Links below (Apache licenses all round too!) ⬇️ 🔗NeRF Studio: 🔗CamP (Camera Preconditioning for Neural Radiance Fields): 🔗Zip-NeRF (Anti-Aliased Grid-Based Neural Radiance Fields): 🔗SMERF vs. Gaussian Splatting video: Enjoyed this thread? You might also enjoy my feed on all things AI, 3D and sPaTiaL CoMpuTiNg: @bilawalsidhu

@nerfstudioteam @GoogleAI So good.

@nerfstudioteam @GoogleAI So good!

@nerfstudioteam @GoogleAI 3d nerds this week

@bilawalsidhu Check DM plz 🙏

@nerfstudioteam @GoogleAI 😯

@nerfstudioteam @GoogleAI Wow, open source 3D capture is really leveling up! 🚀📸

@nerfstudioteam @GoogleAI There is a now an accepted view that these (nerf/splat) scene techniques don’t have to create geometry artifacts (I e. the reconstruction) to be called reconstructions? Without these, they seem like clever novel rendering, but fall (far) short as reconstruction pipelines.
