MIT AI Creates Future Videos From Still Images

Posted: Nov 29 2016, 7:17am CST | by , Updated: Nov 29 2016, 7:19am CST, in News | Latest Science News

 
MIT AI Creates Future Videos From Still Images
Given a still image from a scene, the CSAIL team's deep-learning algorithm can create a brief video that simulates the future of that scene. Image: Carl Vondrick/MIT CSAIL
  • MIT Scientists create Future-Predicting Videos

Certain scientists from MIT have created future-predicting videos that give a premonition of what will come in the forthcoming scenes in a video clip.

Life in a dynamic environment that is physical shows us that we need a thorough understanding of our milieu in order to survive in it. With a minimal amount of thought we can manipulate objects in the environment and handle situations that arise at every step. This all may seem as easy as apple pie, yet in actuality it requires a lot of cognitive processing that has the power of evolution behind it.

This ability comes naturally to us yet for machines to copy this technique, it takes an effort that is monumental and difficult. Making computers and AI equipment recognize how to act in a changing environment is at best a tall order.

Scientists from MIT have made a deep-learning algorithm that creates future scenarios of a video clip based on limited information.

The database contains two million videos which constitute a year’s worth of footage. Future prototypes of this technology could be used in driverless cars and surveillance cameras.

The algorithm can gauge human actions based on past knowledge of the situation at hand. Computers think like humans and thus create prediction patterns that are sifted to select the most likely future trend.

The future scenarios contain visual samples which are gauged and judged based on past examples. The scenes are processed in one go.

It is almost a level of complexity that resembles having a conversation with everyone in the room at a party. It might be impossible but then as they say, anything is possible in cyberspace. The computer models reality with its own heuristics or rules of thumb.

Two competing neural networks are used in this sort of adversarial learning. Video clips of every setting from beaches to train stations and hospitals to golf links have been tested via this method.

There are some commonsensical things which have been left out of this prediction modeling. Yet it will do for now. Collating the information and deducing possible future scenarios from it is a very complex process.

In the future, scenes can be placed under camera surveillance and anything which might happen could be prognosticated based on certain rules of the game.

This story may contain affiliate links.

Find rare products online! Get the free Tracker App now.


Download the free Tracker app now to get in-stock alerts on Pomsies, Oculus Go, SNES Classic and more.

Latest News

Comments

The Author

<a href="/latest_stories/all/all/20" rel="author">Sumayah Aamir</a>
Sumayah Aamir (Google+) has deep experience in analyzing the latest trends.

 

 

Advertisement

comments powered by Disqus