Several of the most compelling radiance field papers and methods have been released by Google over the last year. With each paper release, there's often a clamoring for their release.
In a surprising move, Jon Barron, one of the original authors of NeRF and a Google AI Researcher has released the code for not only Zip-NeRF, but also for CamP.
CamP stands for camera preconditioning and it seeks to have a more precise understanding of a camera's positioning in space. To be honest, I never thought we would see a release of CamP and I am extremely excited to give it a try. It makes a large difference in the reconstruction quality and is more fine tuned than other structure from motion methods, such as COLMAP. I am truly excited about this release and further hope that other methods contemplate bringing CamP into their pipelines. Especially given that it comes with an Apache 2.0 license!
Zip-NeRF holds a special place in my heart because it was the first NeRF based method I received multiple texts from random friends and family members who had seen the viral posts on Reddit and recognized that they were NeRFs. At the time, it was very validating for me to continue pursuing this website, as it was released only a few months into starting to write.
Both Zip-NeRF and CamP are able to be installed through Barron's Github.
Also, as an aside, I went on a date in NYC last night. The reason why I mention that is we were looking at a neighborhood I hadn't spent much time in before, and clicking through each Google Maps page was a bit difficult until I landed on Holywater, which featured a full NeRF in its description. Seeing the entire space in a Luma Flythroughesque manner was extremely helpful. Thanks, Google!
