Olaf Dünkel

ELLIS PhD student with MPI-INF & Uni Oxford

[Olaf Dünkel @olaf_duenkel]

How well do vision foundation models understand the parts and the structure of objects?

Introducing SOCO, our benchmark for Semantic Object Correspondence, appearing as a Spotlight at #ECCV2026.

🌐 https://t.co/qDXYlKQZvS

👇

[Adam Kortylewski @AdamKortylewski]

🚀 We're hiring multiple fully funded PhD students through the #ELLIS PhD Program!

Join us to work on generative AI, world models, and AI safety.

https://t.co/v9V1DCZJ1b

Apply by Oct 31 via https://t.co/GSIsBVzR4y

Spread the word!

[Anates Labs @Anates_Labs]

@odysseyml takes the first place on Physics-IQ Verified in the video to video (v2v) category with Odyssey 3 Pro.

Full Leaderboard: https://t.co/HvkfSlIfsW

In the image to image (i2v) category we still have a head to head race with @bfl_ai.

[David Nordström @davnords]

Today we release Poincar3, our newest 3D model trained entirely through SSL on RGB image sequences.

I find the coolest thing is that, without labels, it learns multi-view geometry. Just using the attention map, reliable tracks can be formed (see below).
https://t.co/P1nShNSMR6

[Vaibhav Agrawal @va1bhavagrawal]

Excited to share PartConcepts, a unified mechanism for fine-grained part localization and generation, accepted to #NeurIPS 2026 @NeurIPSConf 🥳🥳🥳

🌐https://t.co/99u5qyfjP3

[Olaf Dünkel @olaf_duenkel]

🏆 And congratulations again to all authors of "Target-Bench: Can Video World Models Achieve Mapless Path Planning with Semantic Targets?" for winning the Best Paper Award!
Check out their project page: https://t.co/49QSwuskcd

[Olaf Dünkel @olaf_duenkel]

🌀That was World Models in the L∞p at #ECCV2026 !

A huge thanks to our speakers @RanjayKrishna, @vincesitzmann, @lealtaixe , @CarlDoersch , @_amirbar , @taiyasaki

Thank you to all poster presenters and everyone who joined the discussions in Malmö !

[Olaf Dünkel @olaf_duenkel]

🌀That was World Models in the L∞p at #ECCV2026 !

A huge thanks to our speakers @RanjayKrishna, @vincesitzmann, @lealtaixe , @CarlDoersch , @_amirbar , @taiyasaki

Thank you to all poster presenters and everyone who joined the discussions in Malmö !

[Olaf Dünkel @olaf_duenkel]

WOOP @ #ECCV2026 !

As part of the ECCV workshop on World Models in the Loop, we are inviting submissions of papers to our Non-Proceedings track, accessible via https://t.co/QbjzUakIEi

Organizers: @NhiPham49 @KenJSimmoteit @thunguyenphuoc @AdamKortylewski

See you in Malmö!

[Nhi Pham @NhiPham49]

We had a great first half of our workshop ! Big thanks our speakers @RanjayKrishna, @vincesitzmann, @lealtaixe!
2nd half happening now! with talks from @CarlDoersch (16:10-16:40), @_amirbar, (16:40 - 17:10), and a panel at 17:10.

[Olaf Dünkel @olaf_duenkel]

We had a great first half of our workshop !
Looking forward to our 2nd half with talks from
Carl Doersch (16:10-16:40)
Amir Bar (16:40 - 17:10)
and a surely spicy panel discussion with Amir Bar, Carl Doersch, Vincent Sitzmann, Andrea Tagliasacchi (17:10 - 17:50)

[Olaf Dünkel @olaf_duenkel]

WOOP @ #ECCV2026 !

As part of the ECCV workshop on World Models in the Loop, we are inviting submissions of papers to our Non-Proceedings track, accessible via https://t.co/QbjzUakIEi

Organizers: @NhiPham49 @KenJSimmoteit @thunguyenphuoc @AdamKortylewski

See you in Malmö!

[Thu Nguyen-Phuoc @thunguyenphuoc]

A full house for our #ECCV2026 workshop 🌀 World Models in the Loop! More talks and a panel discussion still to come. 📍Join us now at Malmö Arena Hotel, Terrassen!

#worldmodels #WorldModelEvaluation #Evaluation

[Olaf Dünkel @olaf_duenkel]

17:10 - 17:50 : 🎤 Panel discussion with Amir Bar, Carl Doersch, Vincent Sitzmann, Andrea Tagliasacchi,
moderated by Thu Nguyen-Phuoc
17:50 - 18:00: Closing Remarks

[Olaf Dünkel @olaf_duenkel]

Join us at 🌀 𝗪𝗼𝗿𝗹𝗱 𝗠𝗼𝗱𝗲𝗹𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗟𝗼𝗼𝗽 Workshop this afternoon
@eccvconf
📅 Tuesday, 8 September | 13:30 – 18:00 | Malmö Arena Hotel, Terrassen

Schedule:
13:30 - 13:40 : Opening + 🏆 Best Paper Award
13:40 - 14:10 : Ranjay Krishna (UW / Microsoft AI)

[Olaf Dünkel @olaf_duenkel]

WOOP @ #ECCV2026 !

As part of the ECCV workshop on World Models in the Loop, we are inviting submissions of papers to our Non-Proceedings track, accessible via https://t.co/QbjzUakIEi

Organizers: @NhiPham49 @KenJSimmoteit @thunguyenphuoc @AdamKortylewski

See you in Malmö!

[Olaf Dünkel @olaf_duenkel]

14:10 - 14:40 : Vincent Sitzmann (MIT)
14:40 - 15:10 : Laura Leal-Taixé (NVIDIA / University of Toronto)
15:10 - 16:10 : ☕ Poster session + coffee
16:10 - 16:40 : Carl Doersch (Google DeepMind)
16:40 - 17:10 : Amir Bar (Imperial / AMI Labs)

[Olaf Dünkel @olaf_duenkel]

17:10 - 17:50 : 🎤 Panel discussion with Amir Bar, Carl Doersch, Vincent Sitzmann, Andrea Tagliasacchi,
moderated by Thu Nguyen-Phuoc
17:50 - 18:00: Closing Remarks

[Olaf Dünkel @olaf_duenkel]

@RanjayKrishna @vincesitzmann @lealtaixe @CarlDoersch @_amirbar @taiyasaki @eccvconf

[Olaf Dünkel @olaf_duenkel]

🌀 One day to go!

Join WOOP tomorrow from 13:30–18:00 at #ECCV2026.

🎙 Invited talks: @RanjayKrishna, @vincesitzmann, @lealtaixe, @CarlDoersch, @_amirbar

🗣 Panel discussion: @vincesitzmann, @CarlDoersch, @_amirbar, and @taiyasaki

🔗 https://t.co/71Ow3orTK3

[Olaf Dünkel @olaf_duenkel]

WOOP @ #ECCV2026 !

As part of the ECCV workshop on World Models in the Loop, we are inviting submissions of papers to our Non-Proceedings track, accessible via https://t.co/QbjzUakIEi

Organizers: @NhiPham49 @KenJSimmoteit @thunguyenphuoc @AdamKortylewski

See you in Malmö!

[Olaf Dünkel @olaf_duenkel]

@Gerrylwk Examples might be heavy weather changes, impact by external effects like flooding etc.
Essentially, we are targeting not only evaluating point matching but also registration, which can be also relevant for scenes, not only objects.

[Olaf Dünkel @olaf_duenkel]

How well do vision foundation models understand the parts and the structure of objects?

Introducing SOCO, our benchmark for Semantic Object Correspondence, appearing as a Spotlight at #ECCV2026.

🌐 https://t.co/qDXYlKQZvS

👇

[Olaf Dünkel @olaf_duenkel]

@Gerrylwk Our current setting focuses one single objects but extending this formulation to scenes would be surely interesting! Eg finding correspondences across scenes that have similar scene layout / object composition / semantics etc but still vary substantially...?

[Nhi Pham @NhiPham49]

🌀 One day to go!
Join WOOP tomorrow from 13:30–18:00 at @eccvconf.
🎙 Invited talks: @RanjayKrishna, @vincesitzmann, @lealtaixe, @CarlDoersch, @_amirbar
🗣 Panel discussion: @vincesitzmann, @CarlDoersch, @_amirbar, and @taiyasaki
🔗 https://t.co/pQPc6XBO98
#ECCV2026 #WorldModels

[Thu Nguyen-Phuoc @thunguyenphuoc]

🌀 One day to go!
Join WOOP tomorrow from 13:30–18:00 at #ECCV2026.
🎙 Invited talks: @RanjayKrishna, @vincesitzmann, @lealtaixe, @CarlDoersch, @_amirbar
🗣 Panel discussion: @vincesitzmann, @CarlDoersch, @_amirbar, @taiyasaki
🔗 https://t.co/xb7lJ3t92i
#ECCV2026 #WorldModels

[Olaf Dünkel @olaf_duenkel]

Shared work with
@SunagadR (equal contribution)
Haoran Wang, @DavidTHoffmann, Christian Theobalt & @AdamKortylewski

Find us at #ECCV (Thu Sep 10):
🎤 Spotlight: 10:05–10:10
📌 Poster #477: 10:30–12:30
Happy to discuss about correspondence, representations, and model eval!

[Olaf Dünkel @olaf_duenkel]

[5/6]
Alongside SOCO, we also release OmniProbe, a unified framework for evaluating visual representations across:
→ Correspondence
→ Segmentation
→ Depth & surface normals
→ 3D pose
→ Detection
→ Tracking
→ Classification
→ and more
🔧 https://t.co/ICrHWpO0lg