Google NeRF in the Wild 2D Photo 3D Model
H/T: PetaPixel
Google researchers have revealed their updated NeRF-W (Neural Radiance Fields in the Wild) algorithm, which basically creates detailed 3D models of locations around the world by using the reference data from 2D tourists’ photos. This algorithm can also remove unwanted objects and normalize the lighting conditions so they look as natural as possible. Nerf-W is applied to internet photo collections of famous landmarks and the resulting 3D models are even more photo-realistic than in the actual images. Read more for a video and additional information.

Technically speaking, NeRF-W disentangles the shared, underlying 3D geometry from transient objects and photometric variations, thus generating a consistent, photo-realistic scene representation that can be rendered from novel viewpoints. This means that the 3D models are even cleaner and filled with less noise than the neural renderings of the past.

Lenovo IdeaPad 3 15 Laptop, 15.6' HD Display, AMD Ryzen 3 3250U, 4GB RAM, 128GB Storage, AMD Radeon Vega 3 Graphics, Windows 10 in S Mode
1,687 Reviews
Lenovo IdeaPad 3 15 Laptop, 15.6" HD Display, AMD Ryzen 3 3250U, 4GB RAM, 128GB Storage, AMD Radeon Vega 3 Graphics, Windows 10 in S Mode
  • Powered by the latest AMD Ryzen 3 3250U processor with Radeon Vega 3 graphics, the AMD multi-core processing power offers incredible bandwidth for getting more done faster, in several applications at once
  • The 15. 6" HD (1366 x 768) screen with narrow side bezels and Dopoundsy Audio deliver great visuals and crystal-clear sound for your entertainment
  • 128 GB SSD M.2 NVMe storage and 4 GB DDR4 memory; Windows 10 installed
  • Keep your privacy intact with a physical shutter on your webcam for peace of mind when you need it
  • Stay connected: 2x2 Wi-Fi 5 (802. 11 ac/ac(LC)) and Bluetooth 4.1; webcam with microphone; 3 USB ports, HDMI and SD card reader

NeRF-W captures lighting and photometric post-processing in a low-dimensional latent embedding space. Interpolating between two embeddings smoothly captures variation in appearance without affecting 3D geometry,” said the research team.