Google Maps has revolutionized the way we choose venues by introducing Immersive View. This cutting-edge feature combines machine learning and computer vision to create a realistic digital model of the world. Immersive View allows users to explore indoor spaces with interactive, photorealistic, and multi-dimensional experiences, helping them make confident decisions about where to go.
Capturing Venues in Immersive View:
To create high-quality Immersive View reconstructions, Google Maps utilizes neural radiance fields (NeRFs), a state-of-the-art method for 3D reconstruction and novel view synthesis. With the help of professional photographers, Google captures detailed photos of indoor spaces using DSLR cameras within a short timeframe. These photos undergo automatic scanning and blurring to protect privacy. They are then processed using a structure-from-motion pipeline, which solves for each photo’s camera parameters and enables the NeRF reconstruction process.
Enhancing NeRF Reconstruction:
Google employs various techniques to enhance the quality of NeRF reconstructions. They build on the mip-NeRF 360 model, which consistently produces high-quality results. Additionally, they incorporate low-dimensional generative latent optimization (GLO) vectors and exposure conditioning, providing better accuracy in capturing appearance information and controlling the brightness of dimly-lit venues. NeRF models are trained on TPU or GPU accelerators, constantly improving computation efficiency and reconstruction quality.
Delivering an Immersive User Experience:
With a trained NeRF model, Google Maps can generate new photos of a venue from any viewpoint. To deliver a user-friendly experience, Google designed a controllable 360° video player that allows users to navigate through spaces as if flying on a predefined path. By using videos instead of real-time inference or baked representations, Google ensures scalability and provides a seamless exploration experience for users. The video format also offers better editorial control and easier quality inspection.
Google Maps’ Immersive View is revolutionizing the way users choose venues. By combining machine learning and computer vision, Google has created a stunningly realistic digital world that empowers users to explore and make informed decisions. With immersive 360° videos, users can virtually experience spaces and confidently plan their visits. Google continues to innovate and improve its techniques to provide even better reconstructions and user experiences in the future.