It’s interesting how AI has evolved and brought about some pleasant consequences in the society. The consequences have been so stupendous they continue to amaze us. Such is an example of Nvidia using AI to create pretty believable fake videos.
You might be wondering that similar work has been done before, it’s not something new. Of course that bit is true, yet what Nvidia did this time around was to make use of a generative adversarial network, or GAN. These are very good at simulating visual data and work on the basis of neural networks. The networks separately make the data and then acts as a judge letting go of the inaccurate samples.
This way, AI learns over time, creating better visuals over time. And its interesting to note how this kind of software has been widely used in the industry to create all sorts of illusions, from dummy faces to new clothing designs and what not.
“The use of GANs isn’t novel in unsupervised learning, but the Nvidia research produced results — with shadows peeking through thick foliage under partly cloudy skies — far ahead of anything seen before,” the company said.
Nevertheless, the company’s research has a bonus to it; learning under less supervision. What this means is that compared to the other GANS, Nividia’s program requires much less datasets to generate the required data. As Nvidia researcher Ming-Yu Liu explained to The Verge, which means that if you use a GAN for day to night conversion, in order to simulate you have to get it some pairs of images of the same location at day and at night. Examining these images it would then produce new exciting models. Pretty cool, isn’t it?
In their blog post, Nvidia has mentioned the developments in AI and the concept of Deep Learning discuused a the NIPS conference of 2017. The company explained how it has accepted two papers to the conference this year and at the same time mentioned its own contribution of two others. Scrutinizing the team Nvidia has for this AI department, it could be seen that 120 people work in the field and have plenty of resources. This in itself explains the copany’s standing in the field.
The tech just described isn’t only useful for scene transformations, Nvidia believes this technology would come in quite handy for the self driving cars. The data could be collected over time and then simulated to create virtual conditions in weather, covering all the visual bases for vehicles bound to travel in wild.
This would be a huge leap for the self-driving cars. The information gathered and displayed would help maneuver the cars through various terrains in every weather and seasonal conditions.
You can get detailed information regarding the process and the new technology on the Nvidia Blog, and on the dedicated research page.