NVIDIA engineers were able to train the neural network to render a new 3D environment after training on existing videos. The work was carried out by a team of researchers led by Brian Catanzaro (Bryan Catanzaro), vice president of applied deep learning.
According to researchers, this technology will allow developers to create an interactive environment for training robots, autopilots, etc., by training DL models on video from the real world. In parallel with this, costs and time for their creation decrease.
Now the study is demonstrated at the NeurIPS conference in Montreal, in which well-known researchers of artificial intelligence are participating. Engineers have created a simple driving simulator that allows visitors to navigate interactively through the environment created by AI.
News in Russian on