A machine learning algorithm has emerged that can turn a photograph of a landmark into 3D animation. True, for now, AI only needs high-quality images, otherwise, an error may occur.
Researchers from Google have presented detailed 3D scenes of famous landmarks around the world. They used photographs that they downloaded from the Internet and processed them using machine learning.
On the project’s GitHub page, the researchers shared 3D scenes of the Brandenburg Gate in Berlin, the Sacre Coeur Basilica in Paris, and the Trevi Fountain in Rome. Machine learning has created a detailed 3D rendering from this material, in which you can move the view, and the external scene changes with lighting effects.
Our paper, “NeRF in the Wild”, is out! NeRF-W is a method for reconstructing 3D scenes from internet photography. We apply it to the kinds of photos you might take on vacation: tourists, poor lighting, filters, and all. https://t.co/lawLB4eEup (1/n) pic.twitter.com/UuvUm4eVzm
— Daniel Duckworth (@duck) August 6, 2020
The researchers explained that their algorithm can run into difficulties when using images with “unusual parameters.” Photos are not processed correctly in poor light, incorrect exposure, or post-processing. In addition, images from the Internet often contain moving objects – for example, people or cars. They can also interfere with the algorithm.
“Two photographers can stand in one place and take a picture of the same attraction, but the algorithm will process one frame correctly, and spoil the other. We need to refine the AI to make it work correctly, ”the researchers noted.
The resulting model may contain glitches or other errors. Researchers are developing special extensions that soften the processing with AI, which leads to better results.