
Assistant professor @TUEindhoven | Prev. visiting @RWTHVisionLab | Computer vision
Joined June 2016
- Tweets110
- Following442
- Followers240
- Likes2.2K
Daan de Geus retweeted
I think a thread is needed here 🙂. VINO is a DINO-style self-distillation framework for 3D scenes: a student network learns to predict targets from a teacher, which is itself an EMA of the student.
1/10
Replying to @FrancisEngelman @adr_kruse @KadirYilmaz_CV @dcdegeus @BastianLeibe @RWTHVisionLab @USI_INF @TUeindhoven1
Any thread or TLDR?
Daan de Geus retweeted
Excited to share our latest work, VINO! 🍷
We redesigned self-supervised representation learning for 3D scenes with the Volume Transformer and closed the gap to supervised representations 🚀
Need DINO for 3D/volumetric data?
Check out: 🍷VINO: Revisiting 3D Representation Learning with Volume Transformers
📝arxiv.org/abs/2611.07176
🏠adriankruse.github.io/VINO/
Great collab w/ @adr_kruse @KadirYilmaz_CV @dcdegeus @BastianLeibe
@RWTHVisionLab @USI_INF @TUeindhoven1
Daan de Geus retweeted
Need DINO for 3D/volumetric data?
Check out: 🍷VINO: Revisiting 3D Representation Learning with Volume Transformers
📝arxiv.org/abs/2611.07176
🏠adriankruse.github.io/VINO/
Great collab w/ @adr_kruse @KadirYilmaz_CV @dcdegeus @BastianLeibe
@RWTHVisionLab @USI_INF @TUeindhoven1
Daan de Geus retweeted
Out of sight shouldn’t mean out of memory. 👀🧠
Meet LVMT: Video Mask Transformer for Long-term Video Segmentation, accepted at #WACV2027! 🎉
🏆 State-of-the-art video segmentation across six benchmarks
⚡ 10× faster than the prior state of the art
💻 github.com/tue-mps/lvmt
Daan de Geus retweeted
(1/8)
We are excited to share our new model ARROW 🏹 !
ARROW brings D4RT-style decoding to broader input types, including multi-view videos and unordered collections of images.
📜arXiv: arxiv.org/abs/2610.01314
🌐Project page: vision.rwth-aachen.de/arrow
Daan de Geus retweeted
Students sometimes ask me if it still makes sense, in this accelerating age, to pursue a PhD in AI.
Perhaps counterintuitively, I think it's a great time to do so.
I wrote up some thoughts on this here: web.mit.edu/phillipi/www/wri…
Daan de Geus retweeted
Really happy to see SurGe accepted to NeurIPS 🫶
Daan de Geus retweeted
I’m finalizing my ICLR submissions and realizing how much I miss the “old school” way of writing papers.
My favorite part of writing a paper used to be figuring out how to creatively present the results in visual form. My advisor insisted that figures were the most important anchors of a paper: the first thing reviewers would look at, and often the only thing they’d remember. I always start there, and try to fit all of my findings into a visual story, iterating on my figures in lieu of a first draft.
Now we can have LLMs one-shot our graphics into homogeneous oblivion.
I wonder how often people still ask, “What would make this paper even better?”
I’m thinking about how someone with no background in the subject might read it versus how an expert might read it, and what would be in it for either of them. I’m thinking about the simplest, most representative example of the phenomenon I’m trying to measure and characterize, and how to put it right at the beginning so the reader has something concrete to carry through the paper. I’m debating how to introduce notation only insofar as it’s meaningful, instead of adding more clutter, so I can keep referring back to those pointers throughout the paper and ground all sections in a common vocabulary.
I’m noticing related work sections shrink over the years or be demoted to placement just before the conclusion, and wondering how much that reflects a change in what we value. With 60k abstract registrations this year, it’s hard to imagine that someone else isn’t doing the exact same thing, if it hasn’t already been done. And if references only stretch back to a year ago at best, what’s the point of adding context no one is going to care about?
And then I start wondering if I’m overthinking it. Maybe research communication has fundamentally changed. Maybe hardly anyone will read the paper, and I’m better off fitting everything interesting into the abstract. – at Los Angeles, CA
Daan de Geus retweeted
Replying to @KadirYilmaz_CV
@KadirYilmaz_CV presenting Volt - pure Transformer model for 3D scene understanding. #ECCV2026
Going to @eccvconf? 👀
Workshop Radar is ready and online for #ECCV2026 🚀
🔎 Find workshops & tutorials
⭐ Save your favorites
📅 Explore the schedule in one place
Open source. Mobile-friendly. No account required.
Less tab chaos, more conference. 😄
workshopradar.vercel.app
🤖 Made with AI
Daan de Geus retweeted
We are honored that Volt has been selected as one of the 28 Long Orals at ECCV 2026! 🎉
Looking forward to presenting our work and discussing it with others next week in Malmö. Hope to see you there!
Could this be the ViT moment for 3D scene understanding? 🚀
We revisit the good old Transformer architecture and apply it to 3D scene understanding with minimal modifications. #Volt ⚡️
Project page: yilmazkadir.github.io/Volt/
Arxiv link: arxiv.org/abs/2604.19609
(1/5)
Daan de Geus retweeted
*** Deadline extended! ***
New submission deadline: 28 July 2026
If you want to spread the word on your latest research on video understanding, submit a non-archival paper to ECCV's Video4Real workshop!
Daan de Geus retweeted
Are you working on video understanding and you want to share your findings with the community?
Looking for feedback on your ongoing project?
Looking to further advertise your published paper?
Submit a non-archival paper to ECCV's Video4Real workshop! [1/2]
📹 Working on video understanding?
Consider submitting your work to Video4Real 🫎, our #ECCV2026 workshop on Real-World Video Representation Learning!
📆 Important dates:
- Submission deadline: 18 July 2026
- Acceptance notification: 30 July 2026
Join us at the Video4Real workshop at #ECCV2026!
Daan de Geus retweeted
The Video4Real workshop is happening at ECCV 2026!
Submission is now open for non-archival papers on video understanding, and we welcome:
- Recently published papers
- Work-in-progress research
- Extended abstracts
- Datasets, benchmarks, and resources
- Emerging ideas
1/
Daan de Geus retweeted
Volt is accepted to #ECCV2026 🎉 Huge shoutout to my coauthors @adr_kruse, Tristan Höfer, @dcdegeus, and @BastianLeibe.
The arXiv version uses a NeurIPS/ICLR-style template since we find the ECCV template ugly :)
Could this be the ViT moment for 3D scene understanding? 🚀
We revisit the good old Transformer architecture and apply it to 3D scene understanding with minimal modifications. #Volt ⚡️
Project page: yilmazkadir.github.io/Volt/
Arxiv link: arxiv.org/abs/2604.19609
(1/5)























