Back in Amsterdam!

Returning for a long awaited reunion, IBC is back at the RAI in Amsterdam! The content and technology community will connect at this 4 day event, starting September 9, to gain the latest insights on empowering content. Collabora will be present to take part and share in this exciting edition for 2022. If you are planning on attending, please make sure to add booth F57 in Hall 5 to your itinerary & come say hello!

This year, we are showcasing two new Open Source software demonstrations featuring machine learning. The first features a backend video compression that runs on a complete open source stack, and the second is a data quality assurance tool that's designed to integrate into the annotation pipeline.

• Panfrost ML-based video compression

  Traditional video codec techniques are hitting their limit, newer codecs require a 10x increase in computational power for a 2x gain in coding efficiency. It has become clear that radically new approaches are needed. Many have suggested that Deep Learning techniques can yield better results within acceptable computational deadlines. This demo proposes such a technique specifically targeted at videoconferencing.
  
  Using open source software, Collabora has developed an efficient compression pipeline that enables a face video broadcasting system that achieves the same visual quality as the H.264 standard while only using one-tenth of the bandwidth.
  
  In a nutshell, the face video compression algorithms rely on a source frame of the face, a pipeline to extract the important features from a face image, and a generator to reconstruct the face using the extracted and compressed features on the receiving side. We will be demonstrating the entire process, including both encoding and decoding, on a small Rockchip RK3399 board.
  
  

• Analyzing and improving datasets with MLfix

  Training neural networks requires well labelled datasets. Sadly, labelling datasets by humans or by machines is an error-prone process. A leading AI expert Andrew Ng discovered that "In many industries where giant data sets simply don’t exist, I think the focus has to shift from big data to good data." and yet there is a serious lack of tools for working with and understanding datasets. That's why Collabora wrote an Open Source tool called MLfix to analyze and improve the quality of datasets.
  
  The tool works by sorting the images through visual similarity and then showing them in a streamlined user interface. The interface allows images to be marked so that the QA process can be performed. The visual similarity sorting is based on a model trained in an unsupervised manner, so it's not limited to ImageNet-like data.

As a leading contributor to the GStreamer, PipeWire and Wayland projects, Collabora is uniquely positioned to provide the expertise your require to integrate the latest Open Source technologies with the newest multimedia standards. Be it for low latency streaming, cloud transcoding, embedded capture and playback or AR/VR, Collabora can help you with customization, optimization, hardware acceleration and complex multimedia integration so you can build innovative solutions across the digital media space.

Please contact us to set up a meeting with us at IBC.

We look forward to seeing you at the RAI!

Related Posts

Fostering connections at Open Source Summit Latin America

Finding the secret ingredient at Embedded and Kernel Recipes

GStreamer 1.20: Embedded & WebRTC lead the way

Related Posts

Fostering connections at Open Source Summit Latin America

Finding the secret ingredient at Embedded and Kernel Recipes

GStreamer 1.20: Embedded & WebRTC lead the way