Vantage Points
Scientists developed a new artificial intelligence algorithm that can turn simple photos into explorable three-dimensional worlds, New Scientist reported.
The novel neural network distinguishes itself from previous algorithms because it is capable of extracting physical properties from 2D images, explained lead author Darius Rückert.
In their study, Rückert and his team wrote that the neural network requires a selection of images of a scene and a rough 3D model of that scene to generate a world that can be explored from any viewpoint.
In a few instances, they used between 300 and 350 images captured from different angles to produce the perfect 3D worlds. Rückert noted that three-dimensional scenery can also be created with only two images but it wouldn’t be accurate.
“The more images you have, the better the quality,” he said. “The model cannot create stuff it hasn’t seen.”
He is now working to improve the system by having it simulate how light bounces off objects in the scene to reach the camera. The update would require fewer images to create an accurate 3D rendering.
Tim Field, a founder of New York-based company Abound Labs, said the new algorithm is “proof that automated photorealism is possible.”
He suggested that the neural network can be used for generating visual effects in movies and virtual reality walkthroughs of various locations.
Click here to see photos from various vantage points.
No comments:
Post a Comment