An NVIDIA neural network for turning images into 3D objects allows you to “look through the looking glass”
Other examples of AI algorithms are equally impressive
Would you like to feel like Alice through the Looking Glass or a traveler to the Kingdom of Crooked Mirrors? There is such a possibility, thanks only to the neural network.
The fact is that neurons are literally everywhere now – they draw portraits, help with solving problems, develop and, in general, live almost their own lives.
So the NVIDIA neural network came with a surprise as the technology turns pictures into 3D images, but, as it turns out, does not understand that there is a mirror in front, so it creates another space inside it.
At the end of March 2022, NVIDIA introduced the Instant NeRF neural network, which creates 3D objects and even entire three-dimensional spaces from several dozen images. After the company provided the source code of the system to those who wished, they began to share the results of the work of AI algorithms online.
While testing the neural network, developer Hugues Bruyère discovered an unusual “side effect” that occurs when a mirror is in the processed scene. Instant NeRF makes it possible to “fly” into the mirror and even move a little through the “mirror”, while looking around.
In Bruyère’s example, the neural network processed 134 photos. After that, AI algorithms recreated the frames from the missing angles, including reconstructing the reflection in the mirror. According to the developer, the neural network “does not understand” that this is a mirror, but is familiar with the concept of depth.
By the way, the neural network was dubbed the “Polaroid of the future” and is planned to be used in game development. And the fact that there will be another hidden dimension is more of a feature than a bug.