Trending: Adaptive Leadership in Times of Crisis  |  Teacher Insights: 'Lab in a box' projects for home learning  |  Policy Indications: A global collaboration to move AI principles to practice  |  Science Innovations: Translating lost languages using machine learning  |  Science Innovations: Scientists develop ‘mini-brains’ to help robots recognise pain & to self-repair  |  Health Monitor: Ayurvedic Postnatal Care  |  Parent Interventions: Online learning ergonomics: Keep your child engaged and strain-free  |  Parent Interventions: Cow’s milk protein intolerance risk factors   |  Parent Interventions: Safe sports for kids during Covid-19  |  Parent Interventions: E-modules increase provider knowledge related to adverse childhood experiences  |  Technology Inceptions: ICMR validates ‘COVIRAP' by IIT Kharagpur   |  National Edu News: India progressing rapidly towards the goal of indigenously made Supercomputers  |  Best Practices: “Aditi Urja Sanch” Unit at CSIR-NCL, Pune  |  Reflections: What Really Matters  |  Teacher Insights: New Harvard Online course course prepares professionals for a data-driven world  |  
February 24, 2018 Saturday 05:08:49 PM IST

AI could simplify movie production

Technology Inceptions

24th February, 2018: Computer scientists at Duke University in Durham, North Carolina, USA herald a golden age for the movie making with the advent of Artificial Intelligence (AI). The new AI-based technology would enable the production of low budget films and it would consume very short scripts to do the job. The work of the team is considered to be the first text-to- video conversion so far. The team has reported their results recently at a meeting of the Association for the Advancement of Artificial Intelligence in New Orleans, Louisiana.

Here the scientists use, so called “generative” algorithms, which produces images that are correlated to a given series of textual clues. The algorithms coordinates sequence of scenes aligned to the textual suggestions and make those images move in accordance with the text. Such a work has not been attempted earlier the date.

The new algorithms take advantage of machine learning and employ the tools of neural network, and process the given data in a way similar to what human brain does. By large number of trials, the machine learns the art of film making. During the training process, the algorithms assess their own performances and take feedbacks that circulate through millions of network connections. And in this way, they refine their future computations and self-optimize the outcome of the work.

The way generative algorithms work is quite similar to the way an artist may work. In the first stage, it develops a collection of images as directed by the textual clues given. In the second stage, the algorithm refines its work and produces a short video. The training sessions of the algorithms are moderated by a second network, which acts as a “discriminator”, which offers progressively harsher critic as the work proceeds and raises the quality bar for the generator network as the time goes by.


It is a baby-step towards the goal, but a movement in the right direction, believes the scientists. They also predict manifold applications for their methodology. For example, the same technology could be used for better video compressions, using simple text clues, which could be developed into full-fledged movie on-demand. Further, realistic video clips could be used to teach autonomous vehicles to surpass dangerous maneuvers which are critically rare. Similar to any other program that learns deep into the visual world, these algorithms too would find applications into everything from refereeing to surveillance, hope the researchers.Technology Inceptions

 

AI could simplify movie production


 

# Artificial Intelligence (AI) may revolutionize the movie production by creating movies from a few textual clues.

 

24th February, 2018: Computer scientists at Duke University in Durham, North Carolina, USA herald a golden age for the movie making with the advent of Artificial Intelligence (AI). The new AI-based technology would enable the production of low budget films and it would consume very short scripts to do the job. The work of the team is considered to be the first text-to- video conversion so far. The team has reported their results recently at a meeting of the Association for the Advancement of Artificial Intelligence in New Orleans, Louisiana.


Here the scientists use, so called “generative” algorithms, which produces images that are correlated to a given series of textual clues. The algorithms coordinates sequence of scenes aligned to the textual suggestions and make those images move in accordance with the text. Such a work has not been attempted earlier the date.

The new algorithms take advantage of machine learning and employ the tools of neural network, and process the given data in a way similar to what human brain does. By large number of trials, the machine learns the art of film making. During the training process, the algorithms assess their own performances and take feedbacks that circulate through millions of network connections. And in this way, they refine their future computations and self-optimize the outcome of the work.

The way generative algorithms work is quite similar to the way an artist may work. In the first stage, it develops a collection of images as directed by the textual clues given. In the second stage, the algorithm refines its work and produces a short video. The training sessions of the algorithms are moderated by a second network, which acts as a “discriminator”, which offers progressively harsher critic as the work proceeds and raises the quality bar for the generator network as the time goes by.

It is a baby-step towards the goal, but a movement in the right direction, believes the scientists. They also predict manifold applications for their methodology. For example, the same technology could be used for better video compressions, using simple text clues, which could be developed into full-fledged movie on-demand. Further, realistic video clips could be used to teach autonomous vehicles to surpass dangerous maneuvers which are critically rare. Similar to any other program that learns deep into the visual world, these algorithms too would find applications into everything from refereeing to surveillance, hope the researchers.


Comments