A European Innovation Council-backed startup called Deep Render has been making headlines this week with its claim to have developed an artificial intelligence-powered codec that can increase video file compression by up to five times that of existing standards.
The startup says its algorithm has the potential to one day compress files by much more than this too, meaning it will be able to “save the internet” from being overwhelmed by video files. Further, it claims that its compression technique doesn’t impact the quality of video streams. It does this by “exploiting redundancies” in a way that’s more fluid and fine-grained than existing compression algorithms.
Deep Render says its AI algorithm is software-based and can be updated remotely in the devices it's employed in. As a result, it will be able to regularly improve its data compression performance via updates.
Using AI to enhance data compression certainly sounds like a realistic proposal, but it seems likely that Deep Render’s solution will require the hardware that uses it to have some pretty powerful resources, such as an onboard neural processing unit.
Deep Render said it’s currently capable of delivering “5x smaller file sizes” and believes its codec is 80% more powerful than the best existing compression techniques in use today. Ultimately though, it believes its AI algorithm will be able to compress video files by as much as 50-times their current size.
The company is looking to solve a key problem faced by the modern internet. Video streaming is one of the major sources of traffic on the web, so if Deep Render can really compress this data by 50-times, it will free up vast amounts of bandwidth that can be made available for other data sources, such as virtual reality-based metaverses, for example. Deep Render cites data from Cisco, which shows that video accounted for 82% of the world's IP traffic in 2022.
Although that seems like a pretty audacious claim, Deep Render appears to have convinced some heavy hitters that it’s within the realms of possibility. Earlier this year it raised $9 million in funding, with $2.7 million coming from the European Innovation Council. Intel Ventures is also backing the startup. Notably, it has accepted an invitation to participate in Intel’s Ignite accelerator program for startups
If Deep Render continues with its progress, there’s no reason to think it couldn’t supplant existing codecs such as AV1, VVC and HEVC. However, it should be pointed out that Deep Render isn’t the only one that’s looking at how AI can be used to enhance video and data compression.
For instance, a number of companies are exploring ways in which they can enhance the existing VVC (H.266) codec with artificial intelligence capabilities. In Britain, the BBC’s research and development arm is collaborating with the Computer Vision Center iN Barcelona to create neural networks that can boost image compression. The smartphone chipmaker Qualcomm is also said to be working on next-gen AI codecs that will significantly improve data compression.