PhD student Darren Ramsook in the Dept of Electronic and Electrical Engineering, has been awarded first place among academic submissions in the prestigious international Challenge on Learned Image Compression (CLIC). The competition was held during the Digital Compression Conference (DCC) 2024 in Snowbird, Utah, USA. He placed third across all entries including industry submissions with first and second place being taken by teams with members from ByteDance (makers of TikTok) and ZTE (a large producer of video encoders in China).
The competition challenge was to develop a new video compression algorithm to compress video clips down to the ridiculously low bitrate of 50kbps. Entries were judged through a large-scale viewing quality test with real people. This is the highest bar of assessment and Darren’s algorithm was unique in being the only one both to use AI and interoperate with existing standards which makes it applicable now. Being beaten by teams with members from two of the biggest companies in this space is actually an accolade in itself, showing that Universities can still challenge the big guns in AI for video.
To read more about his work, please see here: https://arxiv.org/abs/2401.18021
Example of compression technology which was developed. On the right (C) we show what the image frame should look like. On the left, (A) we have the content compressed with current encoding technologies at a very low data rate of 50 Kb/s. The middle image (B) shows Darren’s algorithm that restores detail to the blurry mess on the left.
Darren's award is a testament to the exceptional research conducted within Trinity College Dublin's School of Engineering. It highlights the ability of our talented students and faculty to compete and excel alongside the biggest names in the field of AI for video compression.
From Left to Right: Fabien Racapé (CLIC Organizer and Senior Scientist @ InterDigital), Darren Ramsook (PhD Student @ TCD), Johannes Ballé (CLIC Organizer and Staff Research Scientist @ Google
Darren Ramsook is part of Signal Processing for Media Applications Group (Sigmedia Group), a leading research group at the confluence of signal processing and artificial intelligence within audio and video applications. His research specifically explores the use of Deep Neural Networks for the perceptual enhancement of compressed videos, addressing the critical challenge of delivering high-quality video content over limited bandwidth. Sigmedia Group was founded in 1998 and is 25 years old. It is a research group within the Department of Electronic and Electrical Engineering and investigates the area of audio and video technology. Sigmedia was founded by Academy Award (Oscar) winner Prof. Anil Kokaram.