Kakao Brain Unveils High-Quality Metaverse Implementation Technology

[Asia Economy Reporter Yuri Choi] Kakao Brain announced on the 20th that it has unified algorithms related to ‘NeRF’ (Neural Radiance Fields) into a single codebase and released it on the largest open-source community GitHub1.

‘NeRF’ is a technology that autonomously learns the color and density of 3D space through deep learning to reconstruct new 3D spaces from 2D images. It especially enhances immersion by vividly reproducing light reflections that vary according to the color and angle of objects. Currently, numerous global tech companies regard ‘NeRF’ technology as one of the core metaverse technologies and are focusing on its development and expansion.

The ‘NeRF-Factory Library’ released by Kakao Brain is a unified code implementation of a total of seven publicly available ‘NeRF’ models to date. Composed of unified code, it not only allows objective comparison of the performance of ‘NeRF’ models but also facilitates technological development and expansion. It was developed through joint research with the research teams of Professors Jominsoo and Park Jaesik at POSTECH and Professor Kwon Inso at KAIST.

Kakao Brain used PyTorch2-based code, a machine learning framework mainly used by AI researchers, while simplifying the code structure to improve accessibility. The company expects that research and development of digital humans, 3D maps, and other high-quality metaverse implementations will continue, using the ‘NeRF-Factory Library’ as a stepping stone.

Kim Ildoo, CEO of Kakao Brain, said, "I hope AI researchers and developers worldwide will utilize the library released by Kakao Brain to expand the base of research on NeRF and neural rendering technologies and apply them to various fields such as high-quality metaverse."