MTS AI Trains Artificial Intelligence to Improve Video Quality and Skip Movie Credits for KION

fgfg Picture

KION Online Cinema has requested MTS AI to come up with ways to make movies and TV shows made in different periods more comfortable to watch and to enable skipping the repeating fragments.

SuperResolution – converting videos to 4K

CV engineers from MTS AI created SuperResolution – an AI-powered technology that allows you to increase image resolution with no detriments to its quality. 4K TV owners to watch old movies and TV series with no image distortion can use it. During an independent survey, the technology developers compared an original 1080p video and a video improved to 4K quality and concluded that the improved video was better in several ways. The 4K video was sharper, with brighter colors, and the image appeared more realistic.

Picture

Automatic credits skipping

The second part of the solution for KION is automatic skipping of movie credits and intros. To enable users to fast-forward the repeating elements of the videos, you would have to label a lot of video data and run it through the ML model to identify segments with credits and intros. If you did this manually, it would take a lot of time, in the first place, and it would also require additional resources, including money, qualified staff and managerial supervision. Seeking to roll out the automatic credits and intros skipping function, MTS AI decided to use neural networks, a set of micro services and NVidia Tesla graphics cards.

The algorithm devised by MTS AI’s engineers involves a number of steps:

  • Decoding an episode – extracting each frame. At this step, we use NVDEC hardware decoders to extract an average of 1,200,000 frames for each season of a series.
  • We run these frames through the neural network without extra labeling while learning (unsupervised learning) to convert images into an array of numbers for subsequent handling.
  • Each frame of one episode is compared against the other frames from all episodes of the season to identify repeating frames. After the decoding process, we get an average of 60,000 frames per episode, and thanks to high-performance graphics cards it only takes us 4 to 5 minutes to compare and identify repeating frames within one season.

All three steps of processing one season take anywhere between 40 and 60 minutes.

What was the result?

  • With the 4K video conversion feature, users can watch movies and TV series with ultimate sharpness and brightness of the image, even if the video was produced in the era of black-and-white cinematography.
  • Users of large screens (computers, add-on modules, Smart TVs) now spend more time watching those TV series that permit skipping credits and intros.
  • Average time that KION users spend on the online platform has also increased.
News
Latest Articles
See more
AI Trends
Sora in Hollywood and AI for Big Football
Team news
MTS AI Signs the Declaration on Responsible Development of GenAI
AI Trends
New Claudes, Singing Portraits and Surgeon’s Avatar
Cases
MTS AI Launches Cloud Video Surveillance for “Gulfstream”
AI Trends
Need for Speed and AI vs. Counterfeit
Media about MTS AI
MTS AI Unveils New LLM Specifically for Business Use
Team news
MTS AI employee joins AI Alliance Science Council
AI Trends
Biothreat Protection and LLM under the Dragon Sign
AI Trends
Sleeper Agents and New Abilities of GPTs
AI Trends
LLM Jailbreak and AI for Training Robots