|
Dima Pozdeev Currently, interested in video pretraining methods and the representations they learn. In the meantime, I am studying Math and Data Science at Technical University of Munich. I received BS in Computer Science at HSE University, where I worked on transfer learning and ensembles with Ekaterina Lobacheva and Dmitry Vetrov. If you'd like to chat or discuss ideas, feel free to reach out on X or via email. I will be available for Machine Learning roles starting October 2026. |
|
Research |
|
Generative Shape Reconstruction with Geometry-Guided Langevin Dynamics
Linus Härenstam-Nielsen, Dmitrii Pozdeev, Thomas Dagès, Nikita Araslanov, Daniel Cremers arXiv, 2026 code / arXiv Traversing diffusion model trajectories while enforcing measurement consistency at each step enables complete and realistic 3D shape reconstruction from partial observations. |
|
DenseMarks: Learning Canonical Embeddings for Human Heads Images via Point Tracks
Dmitrii Pozdeev, Alexey Artemov, Ananta R. Bhattarai, Artem Sevastopolsky ICLR, 2026 project page / arXiv / video / code Learning per-pixel 3D embeddings from in-the-wild head videos improves dense correspondences across subjects and poses, and boosts monocular head tracking. |
|
|
To Stay or Not to Stay in the Pre-train Basin: Insights on Ensembling in Transfer Learning
Ildus Sadrtdinov*, Dmitrii Pozdeev*, Dmitry Vetrov, Ekaterina Lobacheva NeurIPS, 2023 arXiv / X thread / code How to improve ensembles in the transfer learning setup by exploring the target task loss landscape. |