TIGAS
Thin-client interactive Gaussian adaptive streaming. Rasterization lives on a backend; the browser gets view-dependent JPEG frames over QUIC, with an ABR loop aimed at sub-100 ms motion-to-photon latency.
ITEC · CD Laboratory ATHENA · University of Klagenfurt
Systems that put generative models inside the video pipeline.
I am a doctoral researcher working where codecs, networks, and generative models meet: in-painting as compression, semantic instead of pixel streams, and thin-client delivery of interactive 3D Gaussians over HTTP/3.
Research
Conventional streaming still treats every frame as a bag of pixels. My work asks what can be left out of the bitstream if the receiver can reconstruct it: redundant texture for a generative in-painter, a human body that can be driven from keypoints, a 3D scene that only needs the current view rendered on a GPU somewhere else.
That shows up as three threads. ELVIS hides and restores content inside a codec-agnostic pipeline. GenStream replaces pixels with skeleton and camera metadata. TIGAS offloads 3D Gaussian rasterization and streams view-dependent frames over QUIC so a browser can navigate 6DoF scenes without a local GPU.
I also care about the people on the other side of the player. DIGITWISE models individual engagement with digital twins rather than assuming one ABR policy fits every viewer. Related work on voice dubbing and sign-language avatars looks at accessibility as a first-class streaming constraint, not an afterthought.
Before the doctorate I was a data scientist at Bitmovin (root-cause analysis on streaming platforms), a data engineer at Accenture, and I trained as an industrial engineer in Bologna and a data scientist in Milan. The through-line is the same: simplify, prototype, then question the prototype.
Selected systems
GenStream
Semantic streaming for human-centric media: transmit skeletal keypoints, camera pose, and a static 3D background; reconstruct the performer on the client. Extreme compression against HEVC on figure-skating footage, as a sketch of post-codec delivery.
ELVIS
End-to-end learning-based video streaming enhancement: server-side encoding that removes reconstructible regions, client-side generative in-painting that puts them back. Modular across codecs, models, and quality metrics.
DIGITWISE
Digital twins of viewers for adaptive streaming engagement. Per-user XGBoost models predict how far someone will watch, then a unified model can steer provisioning toward features that actually keep people in the session.
PyStream
A Python emulator for evaluating adaptive streaming systems without standing up a full CDN-and-player lab every time. Joint work with Samuel Radler and colleagues at ATHENA.
Practical Game Engineering
A university course on dissecting mechanics and shipping small playable prototypes, digital and analog. Iterative design as an engineering practice, not a vibe.
Publications
Reverse chronological. Author names as on the paper; my name is marked. Corrections welcome at info@. A machine-readable list is in artioli.bib.
2026
arXiv
ACM MM
ACM MM
2025
ACM MM
GenStream: Semantic Streaming Framework for Generative Reconstruction of Human-centric Media
NOSSDAV
End-to-End Learning-based Video Streaming Enhancement Pipeline: A Generative AI Approach
EMS
Real-Time AI-Driven Avatar Generation for Sign Language in HTTP Adaptive Streaming
MHV
Generative AI for Realistic Voice Dubbing Across Languages
2024
MMSys
DIGITWISE: Digital Twin-based Modeling of Adaptive Video Streaming Engagement
MMSys
PyStream: Enhancing Video Streaming Evaluation
MMSys
doctoral
doctoral
Generative AI for HTTP Adaptive Streaming
Teaching & background
Lecturer, Practical Game Engineering, University of Klagenfurt (from 2025). Doctoral research in Computer Science at AAU Klagenfurt (from 2022), in the Institute of Information Technology and the Christian Doppler Laboratory ATHENA.
M.Sc. Data Science, University of Milano-Bicocca (2022). B.Eng. Engineering / Industrial Management, University of Bologna (2019). Industry: Bitmovin, Accenture, Techedge.
Contact
info@emanueleartioli.com is the address for this site, collaboration, and anything that should not go to a university mailbox.
University card: aau.at/team/artioli-emanuele. Code: github.com/emanuele-artioli.