Blog

<p>Throughout 2023, InterDigital&rsquo;s Rennes Office welcomed more than a dozen brilliant interns and researchers to explore and contribute to the industry-shaping research and innovation being conducted at InterDigital.</p>
<p>Before concluding their time at InterDigital, each of the interns reflected on the work they led and experiences they shared working in InterDigital&rsquo;s esteemed labs with world class engineers and inventors.</p>
<p>We thank each of our interns for their insights, their diligence, and the hard work they contributed to the Lab&rsquo;s research and our InterDigital team.</p>
<p style="padding-left: 30px;"><span style="color: #00aeef;"><strong><strong>Baptiste Dumoulin</strong></strong></span>: Evaluating New and Legacy Video/Image Compression Approaches</p>
<p style="padding-left: 30px;"><span style="color: #00aeef;"><strong><strong>Paul Sevellec</strong></strong></span>: Analyzing Neural Network Performances and Properties for In-Loop Filtering</p>
<p style="padding-left: 30px;"><span style="color: #00aeef;"><strong><strong>Th&eacute;o Prieur</strong></strong></span>: Designing Video Encoder Decision Algorithms</p>
<p style="padding-left: 30px;"><span style="color: #00aeef;"><strong><strong>Pieyre Iacone</strong></strong></span>: Implementing Advanced HDR for GPUs</p>
<p style="padding-left: 30px;"><span style="color: #00aeef;"><strong><strong> Mohamed Arbi Ben Youssef</strong></strong></span>: Investigating and Improving LMCS Tuning for Video Encoding</p>
<h3><span style="color: #652e8e;"><strong><strong>Evaluating New and Legacy Video/Image Compression Approaches</strong></strong></span></h3>
<p>While at InterDigital, <em>Baptiste Dumoulin</em> worked to evaluate video and image codecs, specifically new approaches to image/video compression using machine learning, and especially deep learning. These new approaches completely change the traditional paradigm, and it is often difficult to compare new methods with legacy methods. During his internship, Baptiste performed an assessment of both traditional and new codecs in a common framework, allowing our Lab to compare the different methods fairly and determine the consistency of input data, the metrics, assessment, and more.</p>
<p><span style="color: #00aeef;"><strong><strong>Baptiste Dumoulin</strong></strong></span>: &ldquo;During my enriching three-month internship at InterDigital, I had the privilege of collaborating with highly skilled professionals on the fascinating topic of fair comparison between end-to-end (e2e) encoders and traditional ones. The experience was invaluable as I had the chance to explore various methods for E2E compression and deepen my understanding of the VVC (Versatile Video Coding) standard.</p>
<p><strong>&ldquo;I was able to engage in a wide range of rewarding activities, from learning about the state-of-the-art in E2E compression to handling the VVC Test Model (VTM) and running simulations with it</strong>. Working alongside a team of brilliant individuals, I found myself immersed in a supportive and positive environment, for which I am truly grateful to InterDigital and its engineers.&rdquo;</p>
<h3><span style="color: #652e8e;"><strong><strong>Analyzing Neural Network Performances and Properties for In-Loop Filtering</strong></strong></span></h3>
<p>A current trend in the development of next-generation video codecs is the introduction of in-loop filters based on deep neural networks, to enhance the quality of the reconstructed frames in a video. While this approach greatly improves coding performance, the high computational complexity of the neural networks remains a barrier to real world adoption. During his internship, <em>Paul Sevellec</em> analyzed the performances and properties of a neural network for in-loop filtering that was previously proposed in the JVET standardization group. Paul also supported the development of a low complexity machine learning-based component that can be integrated in conventional, or non-neural network, in-loop filters.</p>
<p><span style="color: #00aeef;"><strong><strong>Paul Sevellec</strong></strong></span>: &ldquo;During my internship from March to August, <strong>I had the opportunity to discover the world of video compression in the context of the JVET standardization group. For the first couple months, I developed visualizations and ran tests to extract metrics to have a better understanding of the neural network filtering</strong>. After that, we explored the possibility to develop a hybrid component that would combine a small neural network and an algorithmic part to get the best from both worlds. This was a great experience because I had the opportunity to explore a wide range of ideas and technical aspects in such a great team.&rdquo;</p>
<h3><span style="color: #652e8e;"><strong><strong>Designing Video Encoder Decision Algorithms</strong></strong></span></h3>
<div style="display: flex; align-items: flex-start;">
<div style="flex: 0 0 auto; margin-right: 20px;"><img src="https://www.interdigital.com/resources/img/TheoPrieur.png" alt="" width="288" height="388" />
<div style="text-align: center;"><em>Th&eacute;o Prieur</em></div>
</div>
<div style="flex: 1;">
<p style="word-wrap: break-word;">The Joint Video Expert Team (JVET) of ITU-T and MPEG is currently developing an exploratory video codec called Enhanced Compression Model (ECM) to study video compression algorithms beyond the Versatile Video Coding (VVC) standard. In support, InterDigital&rsquo;s core video compression research team has developed a coding tool that spatially shifts the grid of Coding Tree Units (CTUs) used to divide a picture and compresses it into rectangular block areas. InterDigital&rsquo;s tool increases ECM&rsquo;s compression performances, but still requires a fast encoder decision algorithm to drive this mechanism and make it acceptable by the JVET group.</p>
<p>During his internship, <em>Th&eacute;o Prieur</em> worked on designing video encoder decision algorithms to determine which CTU grid spatial offset to apply when coding and decoding a video picture. Th&eacute;o investigated two types of encoder algorithms: conventional video encoder heuristics and machine learning algorithms. Through his hard work and involvement, he dramatically strengthened his knowledge of video codecs and produced quite promising results around the trade-off between</p>
</div>
</div>
<p>encoder complexity and compression performances when using machine learning-based approaches.</p>
<p><span style="color: #00aeef;"><strong><strong>Th&eacute;o Prieur</strong></strong></span>: &ldquo; Interning at InterDigital has been a wonderful experience so far. I'm enjoying working among a team of experts who are guiding me through my internship and helping me overcome all the difficulties I could face. Thanks to very well detailed documents made by the team I could find everything I needed very quickly. Thus, I was soon up and running. <strong>I learned so much about VVC encoders and artificial intelligence, I'm very grateful to InterDigital for this experience and beyond the technical aspect, I received a lot of valuable advice for my future career.</strong>&rdquo;</p>
<h3><span style="color: #652e8e;"><strong><strong>Implementing Advanced HDR for GPUs</strong></strong></span></h3>
<p>High Dynamic Range (HDR) is a new way to capture, produce, distribute, and render images with increased brightness, deeper contrast, and more colors. InterDigital supports and implements Advanced HDR by Technicolor solutions (AHDR), providing creation and rendering tools to product manufacturers and service providers. During his internship, <em>Pieyre Iacone</em> contributed to the implementation of Advanced HDR by Technicolor algorithms, and successfully developed code with generic interfaces to support different GPU technologies, like OpenGL, OpenCL, and CUDA. Through this project, Pieyre familiarized himself with GPU computing and was exposed to the differences between various GPU programming solutions.</p>
<p><span style="color: #00aeef;"><strong><strong>Pieyre Iacone</strong></strong></span>: &ldquo;My internship at InterDigital has been exciting and a great learning experience. During this internship, I was immersed in a team of highly skilled engineers working on the Advanced HDR by Technicolor (AHDR) solution. <strong>Through my work on AHDR algorithms GPU implementation, I gained expertise in parallel programming (GPGPU) and deepened my understanding of HDR video technologies. Not only did I develop new skills related to my mission, but I also expanded my knowledge in various domains, thanks to the regular scientific seminars provided by employees</strong>. I am truly grateful to InterDigital for providing me with this opportunity.&rdquo;</p>
<h3><span style="color: #652e8e;"><strong><strong>Investigating and Improving LMCS Tuning for Video Encoding</strong></strong></span></h3>
<p>VVC, MPEG&rsquo;s latest video standard, introduced new and different coding technologies. One technology called LMCS, or Luma Mapping and Chroma Scaling, performs a mapping of the input video coding to better exploit the available signal range, however the tuning of this tool is extremely tricky. Throughout his time at InterDigital, <em>Mohamed Arbi Ben Youssef</em> investigated conventional and machine learning-based algorithms to improve the LMCS tuning applied during video encoding and explored normative improvements to the LMCS tool.</p>
<p><span style="color: #00aeef;"><strong><strong>Mohamed Arbi Ben Youssef</strong></strong></span>: &ldquo;During my internship at InterDigital, I delved into the world of video coding, where I had the opportunity to learn about cutting-edge technologies and methodologies. The research aspect of the internship allowed me to explore novel video compression techniques and their applications. I found myself truly passionate about solving complex problems using machine learning and deep learning techniques, which could be pivotal in enhancing video coding efficiency and quality.</p>
<p>&ldquo;Working alongside a highly skilled team, I was exposed to real-world projects that challenged and inspired me. Throughout my time at InterDigital, I was encouraged to think critically and apply theoretical knowledge in a practical setting. <strong>This hands-on experience not only enriched my understanding of video coding and machine learning but also bolstered my problem-solving skills and honed my ability to tackle intricate technical issues</strong>. I am immensely grateful for the opportunity to be a part of InterDigital's dynamic team.&rdquo;</p>