Seven Years of ATHENA: Concluding the Christian Doppler Laboratory

Posted on November 2, 2026 by

The Christian Doppler (CD) Laboratory ATHENA was established in October 2019 to tackle current and future research and deployment challenges of HTTP Adaptive Streaming (HAS) and emerging streaming methods. After seven years, ATHENA officially concluded on September 30, 2026, and we have submitted our final report to the Christian Doppler Research Association (CDG). Time to look back at the achievements of the third and final phase (years 6 and 7; see ATHENA 5-year blog post here) and at what ATHENA accomplished over its entire lifetime.

CD laboratories conduct application-oriented basic research, promote collaboration between universities and companies, and facilitate technology transfer. They are funded through a public-private partnership between companies and the CDG, which in turn receives public funding from the Austrian federal government. ATHENA was supported by Bitmovin as company partner throughout its entire duration.

The big picture

Video streaming remains by far the dominant application on today’s Internet, which makes every improvement in efficiency, quality, and energy consumption relevant for billions of viewers. ATHENA addressed the entire streaming chain in four work packages: content provisioning (WP-1), content delivery (WP-2), content consumption (WP-3), and end-to-end aspects (WP-4). WP-2 was concluded after the first cohort of PhD students completed their theses, allowing us to focus resources on the remaining work packages in the final phase.

If we had to summarize seven years of research in one sentence, it would be this: the trade-off between content complexity, latency, and Quality of Experience (QoE) is not a law of nature, but largely a consequence of static system design. Two publications illustrate this particularly well. The Video Complexity Analyzer (VCA) (ACM MMSys 2022) made content complexity measurable in real time and enabled content-adaptive encoding even for live streaming. ARTEMIS (USENIX NSDI 2024) showed that dynamically adapting the bitrate ladder during a live session reduces encoding computation by 25%, end-to-end latency by 18%, and simultaneously improves QoE by 11%. In other words, all three dimensions can improve at the same time.

Highlights of the final phase

During years 6 and 7, ATHENA contributed 40 papers, including publications in the IEEE Internet of Things Journal, IEEE Transactions on Circuits and Systems for Video Technology, ACM Transactions on Multimedia Computing, Communications, and Applications, IEEE Journal on Emerging and Selected Topics in Circuits and Systems, and ACM Computing Surveys. Content provisioning (WP-1) produced 22 papers, content consumption (WP-3) 8 papers, and end-to-end aspects (WP-4) 10 papers.

Scientifically, the final phase was characterized by a clear shift towards AI and generative methods:

  • Generative AI for streaming: Conventional codecs transmit every pixel, even where the content could be predicted from context. ELVIS (ACM NOSSDAV 2025) removes less important blocks of each frame at the server and lets a generative in-painting model reconstruct them at the client, improving perceptual quality by up to 11 VMAF points over standard encodings at matched bitrates. GenStream (ACM MM 2025, Brave New Ideas) takes this idea further for human-centric content such as sports: instead of pixels, it transmits skeletal poses, camera parameters, and a 3D background model, from which the client reconstructs the video. In first experiments on figure skating footage, the bitrate dropped from 5.8 Mbps to below 10 kbps, a reduction of more than 99.9%, while preserving natural motion.
  • Immersive media: 3D Gaussian Splatting enables photorealistic scenes for virtual and augmented reality, but its models are large and demanding to render. MoQSplat (IEEE MMSP 2026) delivers such scenes progressively over Media over QUIC, eliminating head-of-line blocking and enabling near-instantaneous startup. TIGAS (ACM MM 2026) targets thin clients such as smartphones, smart TVs, and standalone headsets by rendering remotely and streaming over HTTP/3, achieving a motion-to-photon latency of around 60 ms on regional edge nodes, compared with 120 to 300 ms for low-latency DASH/CMAF baselines. A short survey on QoE modeling for volumetric video streaming (IEEE/IFIP NOMS) complements this work with a taxonomy and open research challenges.
  • Energy- and content-aware encoding: ATHENA continued its sustainability line by making energy a first-class encoding objective. LiveESTR (IEEE JETCAS 2025) constructs quality- and energy-aware bitrate ladders for live streaming in real time, reducing the energy required to prepare representations by 74.6% and decoding energy by 29.7% compared with per-title encoding, at only a 2.1% BD-Rate increase. EcoEncode selects codec and preset based on the expected number of views and content complexity, achieving an average BD-Energy reduction of 66.1% for videos with fewer than 2,000 views. DeepQC predicts the quantization parameter needed to reach a target VMAF directly from the source, without pre-encoding, enabling real-time quality-driven encoding on both CPU and GPU infrastructures.
  • AI-based video enhancement: Content-aware super-resolution can improve low-resolution streams by adapting a neural network to each video, but training is costly. EPS (IEEE TCSVT 2026) uses lightweight DCT-based complexity features to select only informative training patches, discarding between 75% and 92% of them while preserving comparable or even better reconstruction quality. Self-Training (IEEE Transactions on Broadcasting 2026) moves content adaptation entirely to the client: the player fine-tunes a super-resolution model using high-resolution segments already in its buffer, eliminating the need to transmit a separate model stream. Under real network traces, this raised average VMAF from 90.84 to 92.99 and reduced quality oscillations, with a negligible increase in rebuffering.

Technology transfer and patents

Over its lifetime, ATHENA resulted in 24 invention disclosures and 21 patent applications (including provisionals), of which 13 US patents have been granted so far together with our company partner. Topics in the final phase include transport-assisted media selection for Media over QUIC, LLM-guided adaptive bitrate algorithms, generative live streaming, and efficient quality control for video encoding.

Several results have been transferred into Bitmovin products and processes, including CAdViSE, the WISH ABR logic in the Bitmovin Player, CO2 calculation for encodings (via the follow-up project GAIA), and VCA-based improvements of per-title encoding. Bitmovin’s live content-adaptive encoding based on real-time VCA achieved around 14.7% bitrate savings at a negligible quality loss in internal validation.

We also continued our commitment to reproducibility through open source tools and datasets. VCA, for example, is used by the Joint Video Experts Team (JVET).

Events and community

A highlight of the final phase was hosting IEEE VCIP 2025 in Klagenfurt in December 2025. In addition, the ATHENA team organized workshops at ACM CoNEXT (EMS 2026), IEEE ICME (P3DMM 2026), ACM MM, and ACM Mile-High Video 2025, two special sessions at IEEE ICIP 2026, and tutorials at ACM MM 2025 and IEEE ICME 2025.

Awards in the final phase include the Best Student Paper Award at the NAB BEIT Conference 2025, the Best Paper Award and the Multimedia Rising Star Award at IEEE MMSP 2026, a top 25 most downloaded article of the IEEE Signal Processing Society, and elevations to IEEE and ACM Senior Member.

People

The most important outcome of a CD laboratory are the people it trains. In the final phase, Dr. Daniele Lorenzi defended his PhD thesis “QoE- and Energy-aware Content Consumption for HTTP Adaptive Streaming” (July 31, 2025), bringing the total number of PhD graduates to eight. Both postdoctoral scholars completed their habilitation:

The next cohort of PhD students is now in the final stage of their doctoral studies and about to submit their theses (working titles):

  • Emanuele Artioli, “From Pixels to Semantics: Compute-Bandwidth Tradeoff in Next-Generation Video Streaming”
  • Zoha Azimi Ourimi, “AI-Assisted Energy-Efficient HTTP Adaptive Video Streaming”
  • Mohammad Ghasempour, “Content-Adaptive Multimedia Coding and Processing for Energy-Efficient Adaptive Streaming”
  • Kamran Qureshi, “Efficient Encoding Strategies for Immersive and Adaptive Streaming Media Applications”
  • Yiying Wei, “Neural Adaptive Video Representation and Generative Reconstruction for HTTP Adaptive Streaming”

ATHENA alumni now work at universities, research institutions, and companies across Europe and beyond, including the University of Klagenfurt, TU Wien, Fraunhofer FOKUS, Yıldız Teknik Üniversitesi, Qualcomm, and Bitmovin. Dr. Hadi Amirpour is now Assistant Professor (tenure track) at the University of Klagenfurt.

What’s next

ATHENA ends, but the research continues. Many of the threads started in the final phase, from generative AI for streaming to immersive media, will be carried forward under the lead of Dr. Farzad Tashtarian, again in close collaboration with Bitmovin. Stay tuned, we look forward to sharing more early next year! In addition, ATHENA’s methods live on in ongoing follow-up projects such as HoloSense, ForestScope, and DORBINE.

Thank you

Thanks to all current and former ATHENA team members: Samira Afzal, Jesús Aguilar Armijo, Hadi Amirpour, Emanuele Artioli, Zoha Azimi Ourimi, Christian Bauer, Alexis Boniface, Ekrem Çetinkaya, Seungmi Choi, Thomas Curry, Reza Ebrahimi, Filipp Eder, Alireza Erfanian, Reza Farahani, Philipp Fößl, Vanessa Fröhlich, Kassian Fugger, Mohammadreza Ghafari, Mohammad Ghanbari (late), Milad Ghanbari, Mohammad Ghasempour, Afshin Gholami, Daniel Glantschnig, Vladislav Golovatenko, Gabriel Gordziejonok, Selina Zoë Haack, MohammadAli Hamidi, Franz Albert Heinrichs, Hermann Hellwagner, Manuel Hoi, Tariqul Islam, Lend Jakupi, Hyewon Jeong, Al·lodi Jutglà Serrat, Luka Kaiser, Georg Kelih, Andreas Kogler, Leon Kordasch, Jonas Kurath, Armin Lachini, Gregor Lammer, David Langmeier, Halime Lezi, Sandro Linder, Daniele Lorenzi, Vignesh V Menon, Sabrina Muhrer, Pranusha Munjuluru, Minh Nguyen, Engin Orhan, Moritz Pecher, Theresa Petschenig, Timo Pfeifer, Leon Prüller, Gabriel Puri, Lingfeng Qu, Kamran Qureshi, Samuel Radler, Raimund Schatz, Yixin Shen, Yahor Siarheyeu, Mykyta Skipenko, Jameson Steiner, Babak Taraghi, Farzad Tashtarian, Per-Luca Thalmann, Leonardo Tocci, Julius van Dillen, Yiying Wei, Yuan Yuan, and Fabio Zinner. Thanks also to the ITEC support staff Martina Steinbacher, Verena Stranner, Nina Stiller, Margit Letter, Marion Taschwer, and Rudolf Messner, and to our international collaborators.

Finally, a big thank you to Bitmovin for seven years of trust and close collaboration, and to the Christian Doppler Research Association for its continuous support and the reviewers for their constructive feedback over the years!

This entry was posted in ATHENA, News. Bookmark the permalink.