Michael Dorkenwald
Student Researcher @ Google DeepMind · PhD Candidate @ UvA

Michael Dorkenwald

Teaching AI models to understand and learn from videos.
I'm a Student Researcher at Google DeepMind and a PhD candidate at the University of Amsterdam, supervised by Yuki Asano and Cees Snoek, as part of the ELLIS PhD program. Video shows how the world changes over time. I work on models that can learn this without supervision, connect it with language, and do so efficiently.

Before that, I got my MSc in Physics from Heidelberg University, working on generative video modeling in Björn Ommer's CompVis group. I also spent time as a research intern at AWS Rekognition and had the pleasure of visiting Kosta Derpanis's lab in Toronto.
Email Google Scholar GitHub CV LinkedIn

News

Research

🔬
Evaluation
Proposing new video-language benchmark
TVBench
💬
Vision-Language
Connecting vision with LLMs
PIN · GLOV
🧠
Representation
Learning from videos without supervision
SIGMA · SCVRL
🎞️
Synthesis
Video generation based on a single frame
cINN · iPOKE
⚡
Efficiency
Making vision models elastic and deployable
Elastic ViTs

Selected Publications

Showing first-author and equal-contribution papers. Full list on Google Scholar. * denotes equal contribution.

Elastic ViTs from Pretrained Models without Retraining

W. Simoncini*, M. Dorkenwald*, T. Blankevoort, C. Snoek, Y. Asano
NeurIPS 2025
ArXiv Project Code

TVBench: Redesigning Video-Language Evaluation

D. Cores*, M. Dorkenwald*, M. Mucientes, C. Snoek, Y. Asano
BMVC 2025
ArXiv Code HuggingFace

SIGMA: Sinkhorn-Guided Masked Video Modeling

M. Salehi*, M. Dorkenwald*, F. Thoker*, E. Gavves, C. Snoek, Y. Asano
ECCV 2024
ArXiv Project Code

PIN: Positional Insert Unlocks Object Localisation Abilities in VLMs

M. Dorkenwald, N. Barazani, C. Snoek*, Y. Asano*
CVPR 2024
ArXiv Project Code

SCVRL: Shuffled Contrastive Video Representation Learning

M. Dorkenwald, F. Xiao, B. Brattoli, J. Tighe, D. Modolo
CVPR 2022 Workshop
ArXiv Project

Stochastic Image-to-Video Synthesis using cINNs

M. Dorkenwald, T. Milbich, A. Blattmann, R. Rombach, K. Derpanis, B. Ommer
CVPR 2021
ArXiv Project Code

Unsupervised Magnification of Posture Deviations Across Subjects

M. Dorkenwald*, U. Büchler*, B. Ommer
CVPR 2020
Paper Project

Experience

2026 — present
Google DeepMind
Student Researcher
Research on efficient video learning.
2022 — present
University of Amsterdam
PhD Candidate — QUVA Lab / ELLIS
Self-supervised video learning, vision-language models, and efficient foundation models. Supervised by Yuki Asano and Cees Snoek.
2022
AWS Rekognition
Research Intern
Self-supervised video representation learning.
2019 — 2022
Heidelberg University
MSc Computational Physics
Research on generative video modeling with Björn Ommer's CompVis group. Research visit at Kosta Derpanis's lab in Toronto.