Fundamentals

What is AI-driven Video Compression?

A recent report published by Sandvine suggests that video now accounts for 65% of all internet traffic, and it’s only growing!

Quoting from the report,

“Demand for video content is soaring, so most platforms are embedding and spreading video within apps to increase views and engagement. Data shows in the first half of 2022, video accounted for a hefty 65.93% of total volume over the internet. That’s a 24% increase over H12021.”

And these unforeseen streaming volumes is precisely where the industry is struggling.

How does a content provider store and stream high-quality video affordably, while providing a great streaming experience to its end-users?

The industry is increasingly under pressure to deliver content at incredibly high resolutions and bitrates, such as 1080p60 fps, 4K, and UHD, for both video-on-demand and live streaming. Furthermore, providers must be able to deliver video at the appropriate bitrates, as buffering and low stream quality will simply drive viewers to its rivals.

The Problems with Traditional Video Compression

The only tool that content providers have in their arsenal to tackle bitrate, filesize, and quality is their video transcoder! But traditional one-size-fits-all transcoders are simply not up to the task when it comes to simultaneously optimizing
• bitrate and filesizes,
• video quality, and
• encoding speeds.

Traditional video compression treats all scenes the same, regardless of how complex they are and this creates problems for content providers.

It is obvious to anyone familiar with video compression that fast-paced videos with lots of action and scene changes need different compression settings than slow-moving videos with mostly dialogue. You wouldn’t apply the same encoder settings for a Simpsons video and an NFL game – would you?

AI-driven video compression can detect different genres and scene complexities
Every genre brings different video complexities, and a video compression system needs to adapt.

Traditional compression can’t adjust for this difference. It either uses too much compression, which reduces video quality but creates smaller files, or not enough compression, which creates larger files that take up more storage and bandwidth.

And what is the impact on video content providers?

• Lower than expected video quality for the specified bitrates.
• Larger file sizes lead to high storage and CDN costs.
• Higher resource costs due to inefficient transcoding.
• Poor customer satisfaction due to buffering, poor video quality, stalls, and large downloads.

• Constant tuning of the encoders to ensure that they produce optimal results for every genre of video.

AI-driven Video Compression

Luckily, there is a better way to compress videos, and that is by using AI in the video compression pipeline.

The use of AI tackles the limitations of traditional compression by focusing on both encoding flexibility and compression efficiency.

And, unlike traditional methods, AI-driven video compression can adjust for each video scene. It analyzes each frame and automatically sets the best compression level, leading to the most optimal compression efficiencies (for both highly complex and simple scenes).

Benefits of AI-driven Video Compression

The advantages of AI when using video compression are substantial!

Cutting costs in half: With traditional methods, storing and transmitting 4K and UHD content can cause storage and bandwidth costs to skyrocket. However, AI-driven video compression can cut video size in half without sacrificing quality, so you can save massive amounts of money on lower CDN costs, faster downloads for users, and less buffering.

Better User Engagement: Using AI in video compression leads to small file sizes (often 30% or more), which means that they are faster to download over the internet, start playing faster on the device, and provide a better user experience. This results in longer watch times and higher ad revenues.

Expand and adapt easily: AI compression can be used to scale up as your video library grows, users’ needs change over time. AI is smart enough to recognize and adopt different video types (sports vs. dramas), different network conditions, etc., and provide the best customer experience. You do not need to deploy your resources to tune the encoder all the time when you see some new genre coming in!

How does Visionular use AI in Video Compression?

Our sustained R&D into AI and ML means that we can encode video in a smarter and faster way using AV1, HEVC, and H.264/AVC video coding standards.

And here is how we used AI in video compression. 

Step 1: Categorise videos and scene types: Our CNN models are best tuned to understand and analyze your videos (your videos can be news, sports, cartoons, or anything else) as they enter the encoder and use the passed results to classify the genre and complexity of your videos. This is our initial starting point for the next phases of tailored compression.

Step 2: AI-Enhancement: We leverage AI enhancement to increase the visual fidelity and enable much lower bitrates without compromising quality. For example, we do a lot of image repair to correct compression artifacts, super resolution to give you extra detail, and intelligent tone mapping to present you with a more cinematic image.

Step 3: ROI-based video compression: Visionular’s AI allocates more bits to the most important regions of the frame. More advanced rate control logic is applied to each region of the frame based on Region of Interest (ROI) and frame-level adaptation to achieve more bandwidth efficient solutions.

A proprietary, in-house AI-powered quality assessment mechanism governs video quality throughout the encoding process for all these services. 

The above technology allows our encoders to

  • reduce bitrates by 30 – 50%
  • deliver high video quality.
  • conserve transcoding resources by turning on/off video coding tools depending on the genre and video complexity. This allows for fast and CPU-efficient encoding, and results in significantly faster encoding times as compared to traditional methods.

Last but not least, our encoders are compliant with industry standards, meaning that the bit stream they generate can be decoded by all the top decoders currently in use (be it on a STB, browser, TV, mobile chipset, software decoder, etc.), enabling enterprises to go into production right away and take advantage of smaller file sizes, lower bitrates, and great video quality.

Check out the video below that shows the improvement in video quality and reduction in file-size and bitrate after using AI in the H.264/AVC video codec.

The Future is Here!

For video streaming businesses, AI compression isn’t just a fad, it’s a strategic advantage. In a competitive market, delivering high-quality videos efficiently is essential. AI compression makes both possible, giving businesses a clear edge.

You can start a no-risk, free trial of Visionular’s AI-driven video compression on the cloud today and experience the power of AI-driven video compression for yourself!

Related Posts