Adilbek Karmanov
I am an upcoming PhD student at the University of Tübingen, advised by Prof. Gerard Pons-Moll in the Real Virtual Humans group, as part of IMPRS-IS. I will work on synthesizing how humans interact with objects and scenes, and on transferring these behaviors to humanoid robots.
Before that, I was a Research Engineer at the Metaverse Center, MBZUAI with Prof. Hao Li, working on 3D avatar synthesis, 3D reconstruction, and neural rendering, including real-time systems for VR and large LED displays.
I received my M.Sc. in Computer Vision from MBZUAI, advised by Prof. Shijian Lu, with a thesis on test-time adaptation in vision-language models. I earned my B.Sc. from Suleyman Demirel University, with an exchange term at UNIST.
I'm always open to discussing research, new ideas, or potential collaborations, so don't hesitate to reach out by email!
Publications
-
SOAP: Style-Omniscient Animatable PortraitsSIGGRAPH 2025
SOAP turns a single stylized 2D photo into a fully rigged, high-resolution 3D avatar with consistent geometry, texture, and animation-ready features.
-
DiffPortrait360: Consistent Portrait Diffusion for 360° View SynthesisCVPR 2025
Diffportrait360 generates fully consistent 360° head views from a single portrait, handling humans, stylized characters, and anthropomorphic forms with accessories.
-
VOODOO XP: Expressive One-Shot Head Reenactment for VR TelepresenceSIGGRAPH Asia 2024 (Journal Track)
VOODOO XP enables real-time, 3D-aware head reenactment from a single photo and any driver video, delivering expressive, identity-preserving, and view-consistent avatars for immersive VR telepresence.
-
Efficient Test-Time Adaptation of Vision-Language ModelsCVPR 2024TDA is a training-free dynamic adapter that enables efficient test-time adaptation of vision-language models through progressive cache updates and negative pseudo labeling, achieving state-of-the-art results without backpropagation.
Experience
Demos
-
VOODOO VR: One-Shot Neural Avatars for Virtual RealitySIGGRAPH 2024 Real-Time Live!
We present a complete solution for real-time immersive face-to-face communication using VR headsets and photorealistic neural head avatars generated instantly from a single photo.
