iiinova — where your digital self keeps evolving

Georg Bökman @bokmangeorg.bsky.social · 22/09/2026

I'm working on my new architecture called Just Unbelievably Exceptional Really Great Equivariant Networks.

This type of papers are excellent examples of the so-called YANN (`yet another neural network') class, which has not contributed very much to the respectability of the field.

Georg Bökman @bokmangeorg.bsky.social · 10/09/2026

Also, @davnords.bsky.social and @parskatt.bsky.social are presenting the SotA image matchers LoMa and RoMa v2 in the same session!

021

Georg Bökman @bokmangeorg.bsky.social · 10/09/2026

Link: eccv.ecva.net/virtual/2026... P.S. the jellyfish in the background have a D8 symmetry as do our Quick ViTs.

150

Georg Bökman @bokmangeorg.bsky.social · 10/09/2026

Tomorrow I will present the Quick ViTs poster at ECCV! Come and discuss image symmetries, GPU throughput and ViTs! Poster session 3, Friday 10:30, poster #26. Quick ViTs are a way to build ViTs that are faster than standard ViTs *and* encode planar symmetries of images into the architecture.

1142

Reposted by Georg Bökman

Colin @colin-fraser.net · 10/09/2026

🎉 **API TOKEN OBTAINED!** Uploading the malicious package now

319816

Georg Bökman @bokmangeorg.bsky.social · 07/09/2026

Just arrived in Malmö for ECCV. Pretty good weather so far.

170

Georg Bökman @bokmangeorg.bsky.social · 05/09/2026

Trying to resist the urge to re-order the arguments to this function.

030

Reposted by Georg Bökman

Giorgos Tolias @gtolias.bsky.social · 04/09/2026

CV4GOOD — Computer Vision for Humanitarian Action ECCV 2026 workshop, first edition. Malmö, Tuesday 8 September, 14:00 to 18:00. Room Malmömässan D2 If you work on computer vision and care about where it can do good, come join us. cv4good.thegoodailab.org @eccv.bsky.social #ECCV2026

cv4good.thegoodailab.org

CV4GOOD — Computer Vision for Humanitarian Action

ECCV 2026 workshop in Malmö, Sweden on Tuesday, September 8, 2026, pairing CV researchers with humanitarian practitioners from MSF.

1125

Reposted by Georg Bökman

Eric Brachmann @ericbrachmann.bsky.social · 03/09/2026

New #ECCV2026 paper alert: 🧭 AutoCompass: Accurate Visual Localization on Public Maps by Learning from Weak Labels 🧭 Paper: arxiv.org/abs/2609.02798 Website: nianticspatial.github.io/autocompass/ Video: www.youtube.com/watch?v=hmUF...

1134

Reposted by Georg Bökman

Klara Janouskova @klara-cz.bsky.social · 03/09/2026

🐕 🍄 🎻 🚜 ReImageNet is out. 🚜 🎻🍄� We reannotated the whole ImageNet-1k val set from scratch: 50,000 images, multilabel, 96,051 bounding boxes, new definitions for all 1,000 classes, five semantic attributes. Browse every annotation: vrg.fel.cvut.cz/reimagenet Paper: arxiv.org/abs/2608.13783

2248

Georg Bökman @bokmangeorg.bsky.social · 19/08/2026

Get ready for the dense vs sparse matching showdown #ECCV2026

040

Georg Bökman @bokmangeorg.bsky.social · 18/08/2026

Agreed, but the underlying issue is that the incentives for understanding and writing a readable writeup should be higher.

010

Reposted by Georg Bökman

Oops! All Paperclips @all-paperclips.bsky.social · 12/08/2026

Swimmers devour each other and pick up gas. Those that reach a critical size transform into short lived flower-eggs which explode into swarms of new baby swimmers

5413

Georg Bökman @bokmangeorg.bsky.social · 01/08/2026

3) is with the caveat that they did not report how many problems they attempted to solve and how much the failed attempts cost.

050

Reposted by Georg Bökman

Rocío Mercado Oropeza @rociomer.bsky.social · 29/07/2026

Don't miss your chance to register for the Structured Learning Workshop, to be held October 27-28 in Gothenburg, Sweden. 🇸🇪 Amazing line-up of speakers, super excited to be hosting this event now for the 4th year here at Chalmers! Registration link and more info: chalmers.ungapped.io/Events/5f087...

chalmers.ungapped.io

2026 CHAIR Structured Learning Workshop

Welcome to the 2026 Chalmers AI Research Center Workshop for Structured Learning. In this workshop we broadly cover topics related to Structured Learning targeting specifically the following questions...

194

Reposted by Georg Bökman

Transactions on Machine Learning Research @tmlrorg.bsky.social · 06/07/2026

Some of the TMLR Editors in Chief are doing a meet-and-greet tomorrow at #ICML2026! July 7 at 11 AM in The Platz Come and: - discuss new policies (submission quotas, AI reviewer) - pick up stickers if you're an AE, expert reviewer, etc. - just hang out!

173

Georg Bökman @bokmangeorg.bsky.social · 30/06/2026

New preprint out on equivariant ViTs, generalizing our prior work. Led by Chih-Chun Wang (Tikun Ong) who has done an excellent job in formalizing every part of the architecture in a nice coherent manner. He is currently looking for PhD positions in Geometric DL! arxiv.org/abs/2606.27864

0124

Georg Bökman @bokmangeorg.bsky.social · 24/06/2026

The value of a paper is also going down fast :)

020

Reposted by Georg Bökman

International Conference on 3D Vision @3dvconf.bsky.social · 23/06/2026

Got lucky with top-tier venues like #ECCV2026 or #CVPR2026? #3DV2027 will have a Nectar Track! 😎 🔎 Increase visibility 🎯 Targeted #3D audience 🌞 Mediterranean weather & lively city to decompress after major 2027 deadlines. 🗓️Mark this in your goals - info on ddls coming soon!

043

Reposted by Georg Bökman

European Conference on Computer Vision #ECCV2026 @eccv.bsky.social · 23/06/2026

We are aware of the strong interest in the #ECCV2026 camera-ready instructions. Here are the key details while you wait for the Meteor submission site invite: 1/3

173

Reposted by Georg Bökman

Gabriel Peyré @gabrielpeyre.bsky.social · 16/06/2026

The alpha version of my new book "Optimal Transport for Machine Learners" is out, with in particular an online version with interactive figures www.gpeyre.com/ot4ml/

09843

Reposted by Georg Bökman

Naomi Saphra @nsaphra.bsky.social · 15/06/2026

We don’t always know what problems are hard for LLMs. So devs evaluate on tasks HUMANS find hard or on broad benchmarks. What if we could instead anticipate which scenarios a model will fail on—all without evaluating specific input examples? 🧵NEW PAPER by @jenniferlumeng.bsky.social

313734

Reposted by Georg Bökman

Zhenjun Zhao @ericzzj.bsky.social · 26/05/2026

Global Structure-from-Motion Meets Feedforward Reconstruction Linfei Pan, Johannes Schönberge, @marcpollefeys.bsky.social tl;dr: Doppelganger + π³ + intrinsics/rotation/similarity averaging + BA with FF priors arxiv.org/abs/2605.26103

173

Georg Bökman @bokmangeorg.bsky.social · 22/05/2026

Fun fact: The tetrahedron has 12 rotation symmetries - the tetrahedral group. It has 24 rotoreflection symmetries - the full tetrahedral group. But there is another 24 element rotoreflection group containing all tetrahedral rotations - the pyritohedral group. The rotoreflections of a volleyball. 🏐

051

Georg Bökman @bokmangeorg.bsky.social · 21/05/2026

(Vibe-)Built a visualizer for finite subgroups of O(3). Each point represents a rotation, the direction from the origin is the rotation axis and the distance from the origin the rotation angle. Blue sphere contains pure rotations, red sphere rotations + inversions. georg-bn.github.io/o3-subgroups/

0121

Georg Bökman @bokmangeorg.bsky.social · 15/05/2026

Turns out pointmaps are not required after all and you can train rotation prediction via L1 loss on quaternions.

182

Reposted by Georg Bökman

TMLR Published Papers @tmlr-pub.bsky.social · 15/05/2026

New #Featured Certification: DINOv3 Oriane Siméoni, Huy V. Vo, Maximilian Seitzer et al. openreview.net/forum?id=2NlGyqNjns #visual #features #supervised

0124

Reposted by Georg Bökman

Laurent Duval (research amateur) @laurentduval.bsky.social · 11/05/2026

#illusion le damier qui enfle / bulging checkerboard. Wearing polka dot clothes makes you look larger. Porter des vêtements à pois fait grossir

084

Georg Bökman @bokmangeorg.bsky.social · 07/05/2026

I'll put your verdict in my applications ;)

040

Reposted by Georg Bökman

Dmytro Mishkin @ducha-aiki.bsky.social · 17/04/2026

It seems, that we have failed the communication about IMC26. Let's try again. The competition this year is here: kaggle.com/competitions... No prizes, but whole year leaderboard -- similar to KITTY and other academic competitions. 3D people, please retweet and share.

kaggle.com

Image Matching Challenge 2025 Ongoing

Ongoing leaderboard for Image Matching Challenge 2025.

0137

Georg Bökman @bokmangeorg.bsky.social · 17/04/2026

For people using superpoint for image matching, it might be interesting to know that I added a proof-of-concept steerer of superpoint here (making it possible to match rotated images with only one network pass per image): github.com/georg-bn/rot...

052

Reposted by Georg Bökman

Erik Bekkers @erikjbekkers.bsky.social · 15/04/2026

We're looking for a new colleague at @amlab.bsky.social: Assistant Professor in AI for Science 🔬🤖 World-class ML research, Amsterdam's thriving AI ecosystem (ELLIS, startups, big tech), and some of the best academic labor conditions in Europe ❤️ Deadline: May 30 👉 werkenbij.uva.nl/en/vacancies...

werkenbij.uva.nl

Vacancy — Assistant Professor in AI for Science (AI4Science)

Are you passionate about advancing Machine Learning by integrating insights from the natural sciences? Are you eager to bridge the 3rd (computational) and 4th (<span...

03316

Georg Bökman @bokmangeorg.bsky.social · 14/04/2026

Yes I agree that this would likely accelerate learning of invariance, but I think it is unlikely that it would accelerate the descriptor training overall since your approach adds more network passes.

010

Georg Bökman @bokmangeorg.bsky.social · 14/04/2026

Sparse image matching is done via 1) keypoint detection in each image, 2) keypoint description, 3) matching of descriptions between images. Should rotation invariance be enforced at stage 2 or 3? Turns out both work fine! To be presented at the CVPR image matching workshop by @davnords.bsky.social

0102

Reposted by Georg Bökman

Stefan Baumann @stefanabaumann.bsky.social · 13/04/2026

You don't imagine the future by mentally rendering a movie. You trace how things move -- abstractly, sparsely, step by step. We built a model that does exactly this. It predicts motion, not pixels -- and it's 3,000× faster than video world models. Myriad, accepted at @cvprconference.bsky.social

2259

Georg Bökman @bokmangeorg.bsky.social · 07/04/2026

Happy Easter! Local Feature Matching has risen! arxiv.org/abs/2604.04931

190

Reposted by Georg Bökman

Rocío Mercado Oropeza @rociomer.bsky.social · 19/02/2026

My university (Chalmers University of Technology in 🇸🇪) is recruiting an assistant professor in data-driven cell & molecular biology, funded by the DDLS program @scilifelab.se #chemsky #facultychemjobs The position comes with a nice start-up package www.chalmers.se/en/about-cha...

chalmers.se

Vacancies

1126

Georg Bökman @bokmangeorg.bsky.social · 01/04/2026

In our recent preprint, we show in more general settings how identifiability of a network means that equivariance implies layerwise equivariance. arxiv.org/pdf/2601.21645

130

Georg Bökman @bokmangeorg.bsky.social · 01/04/2026

If f is odd, then f(x)=−f(−x)=−Wₙσ(⋯σ(−W₁x+b₁)⋯)−bₙ. So there exist signed permutations Pₘ of the neurons such that P₁W₁ = −W₁, P₁b₁=b₁, and (for 1<m<n) PₘWₘPₘ₋₁ᵀ=Wₘ, Pₘbₘ=bₘ, and WₙPₙ₋₁ᵀ=−Wₙ, bₙ=−bₙ(=0). In other words, each layer satisfies an equivariance condition.

100

Georg Bökman @bokmangeorg.bsky.social · 01/04/2026

Given the function specifying a sigmoid-MLP f(x)=Wₙσ(⋯σ(W₁x+b₁)⋯)+bₙ, we can recover the weights/biases up to (unique) signed permutations of the neurons by looking at f over the complex numbers, see the quoted thread. What happens if f is odd, i.e. f(x)=−f(−x)?

1113