InterDigital and Silo AI Collaborate to Build a Platform to Accelerate the Development of AI-Based Video Compression for Machine (VCM) Tools

InterDigital, a leading research and innovation company that develops and licenses foundational wireless and video technologies for mobile devices, networks, and services worldwide, has teamed up with Silo AI, the largest private AI lab from the Nordics, to build a software framework to test and develop AI-based video compression technologies developed by InterDigital’s Labs.

The Research and Innovation Labs at InterDigital lead the way for video innovation and standards, and they develop and license the building blocks for video compression standards which are ubiquitous in our daily life because they are deployed in nearly all video capable consumer devices. InterDigital is continuously developing cutting-edge technologies, placing AI at the core of their research and innovation.

Silo AI supports InterDigital in achieving their industry impact by developing the CompressAI-Vision program suite, a software for image- and video-compression scientists to test, evaluate and train AI-based compression methods. Video Compression for Machines (VCM) research requires the management of large image- and video datasets and the building of complex evaluation pipelines. 

InterDigital is a firm believer in and contributor to cross-company industry collaboration, ensuring a continuously evolving ecosystem where companies can benefit from each other while smaller players have the opportunity to contribute. For this reason, CompressAI-Vision was recently released as an OpenSource Software, and the feedback so far has been positive. The tool frees the capacity of video researchers for other, more complicated tasks. InterDigital's belief in industry collaboration, augmenting its own strong technical capabilities with world-class AI expertise from Silo AI, has been a powerful combination resulting in an innovative computer vision tool beneficial for the whole industry.

Jaideep Chandrashekar, Senior Director, Head of Emerging Technologies Lab at InterDigital, says: 

"Silo AI's engineers worked very closely with InterDigital’s Emerging Technology Lab to translate research ideas into tools that take advantage of cutting-edge AI technologies, helping us deliver the CompressAI-Vision on time." 

AI-based Compression Methods for Scalable Video Research

The AI-based compression methods have recently demonstrated their superiority to traditional image compression methods like jpeg and state-of-the-art video compression codecs used in still picture modes, such as H.265/HEVC. Instead of human vision, compression methods can also be optimized for computer vision tasks, such as object detection and segmentation from images. However, the evaluation pipelines in that context can become quite complex. Typically, in addition to compression and decompression, they include pre- and post-processing steps such as cropping, rescaling, framerate adaptation, and more. The final decoded content is then passed through a neural network-based detection algorithm. The performance of the whole pipeline must be assessed with standard evaluation protocols, while the evaluation must be done using several image-, and video datasets, each including thousands of pictures and up to hundreds of video clips. Calculations are typically done in parallel at large GPU clusters. CompressAI-Vision is a tool that facilitates the construction of evaluation pipelines, the evaluation of custom image compression models, and the management of large image- and video datasets.

Peter Sarlin, CEO & Co-founder at Silo AI, says:

"The work with InterDigital is an excellent example of putting AI at the core of a digital platform. The AI-based tool we’ve built enables more efficient use of advanced technologies, making it well-aligned with our vision of building human-centric AI that frees researchers to do meaningful work."