iiinova — where your digital self keeps evolving

Eric Brachmann @ericbrachmann.bsky.social · 25/09/2026

One of our NeurIPS reviewers thought our "value [...] is real and well-supported". 🤔 🙈

static.klipy.com

I know what you did

ALT: I know what you did

061

Reposted by Eric Brachmann

Lia @liasae.bsky.social · 24/09/2026

I am now keeping this on speed-dial

Edited old IBM printout in black caps on off-white paper.
'A computer can never be held accountable
Therefore a computer must always do the hacking haha what are they gonna do, put a computer in jail? Haha "oops it's alive" lmao'

81100322

Eric Brachmann @ericbrachmann.bsky.social · 24/09/2026

Can I at least change the pod caster persona from "overhyped marketing person" to "cynical scientist"?

020

Eric Brachmann @ericbrachmann.bsky.social · 24/09/2026

What fresh hell is this? Scholar inbox now comes with 2 min podcasts of each paper... I randomly choose two relocalization papers and they both started with "Imagine navigating a large, dark warehouse" ... No - I do not want to imagine this. Was this trained on YouTube slasher flicks? Why "dark"?!

181

Reposted by Eric Brachmann

Angela Dai @adai.bsky.social · 18/09/2026

🚀We have PhD openings in my lab! 3D got pretty interesting :) Generation is getting better, which makes real scenes more interesting: things get used, opened, moved, etc We want to understand that in 3D: function, dynamics, physics 👉apply 3dunderstanding.org/openings.html or via ELLIS

0125

Eric Brachmann @ericbrachmann.bsky.social · 16/09/2026

The plot says "in August", so for a month. The top line just crosses 1000. There were 614 paper in cs CV... yesterday. 🤪 arxiv.org/list/cs.CV/r...

arxiv.org

Computer Vision and Pattern Recognition

020

Eric Brachmann @ericbrachmann.bsky.social · 16/09/2026

In case it's useful for anyone, this plot shows "math" categories only. So CV in there is not "computer vision".

030

Eric Brachmann @ericbrachmann.bsky.social · 15/09/2026

One of the seven millennial problems.

050

Eric Brachmann @ericbrachmann.bsky.social · 14/09/2026

People are busy generating the next 10 paper-shaped objects. No time to spend more effort on stuff that was already accepted.

030

Eric Brachmann @ericbrachmann.bsky.social · 07/09/2026

We also co-organize the Workshop on Recovering 6D Object Pose on Sep 8 (Tuesday) PM. There, we will present the new BOP-Refer benchmark, which tests the capabilities of VLMs for visual grounding. cmp.felk.cvut.cz/sixd/worksho...

cmp.felk.cvut.cz

11th Workshop on Recovering 6D Object Pose (R6D)

11th Workshop on Recovering 6D Object Pose (R6D), organized at ECCV 2026 in Malmö, Sweden.

020

Eric Brachmann @ericbrachmann.bsky.social · 07/09/2026

I will give a talk about scene coordinate regression at the Privacy-Preserving Visual Localization and Mapping Workshop on Sep 8 (Tuesday) around 5:00 PM. sites.google.com/view/ppvlm20...

sites.google.com

PPVLM

Invited Speakers

110

Eric Brachmann @ericbrachmann.bsky.social · 07/09/2026

En route to #ECCV2026! We will present AutoCompass at the main conference 10:30 AM on Sep 11 (Friday) and the Instance-Level Recognition and Generation Workshop as an invited poster, 10:00 AM on Sep 8 (Tuesday). nianticspatial.github.io/autocompass/

nianticspatial.github.io

AutoCompass: Accurate Visual Localization on Public Maps by Learning from Weak Labels

Robust visual localization from noisy GPS labels and accurate relative poses.

190

Reposted by Eric Brachmann

Zhenjun Zhao @ericzzj.bsky.social · 05/09/2026

AutoCompass: Accurate Visual Localization on Public Maps by Learning from Weak Labels Javier Tirado-Garín, @asavio.bsky.social, Shuai Chen, @axelbarroso.bsky.social, Tommaso Cavallari, @dantkz.bsky.social, Victor Adrian Prisacariu, @ericbrachmann.bsky.social arxiv.org/abs/2609.02798

161

Reposted by Eric Brachmann

Evangelos Kazakos @ekazakos.bsky.social · 02/09/2026

Welcome to the ECCV 2026 feed! Post and discuss ECCV 2026 papers, workshops, tutorials, talks, and social events. Use #ECCV2026 to make sure your post appears (other variants work too, e.g. #eccv). Link: bsky.app/profile/ekaz.... Please pin 📌 and share! Thanks! Made by @ekazakos.bsky.social.

295

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

Work done during a stellar internship of Javier Tirado Garín at Niantic Spatial, Inc. With Alan Savio Paul, Shuai Chen, Axel Barroso Laguna @axelbarroso.bsky.social, Tommaso Cavallari, Daniyar Turmukhambetov @dantkz.bsky.social , and Victor Adrian Prisacariu.

000

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

We will present the paper at hashtag #ECCV2026: #116 at 10:30 AM on Sep 11 (Friday) in Poster Session 3. If you prefer a more intimate setting, we will also present it as an invited poster at the Instance-Level Recognition and Generation Workshop, 10:00 AM on Sep 8 (Tuesday).

100

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

The headline numbers of AutoCompass are just amazing and set a new SOTA for this task. By switching to weak supervision, you beat associated baselines. Also some interesting findings: train on coarse position without heading labels only, and accurate heading prediction just comes as a side product.

100

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

So we train with weak supervision instead. The result is a set of losses catering to different training settings: Got images with GPS but no heading? Check. Relative poses but not scale-metric? Sure. Scale-metric relative poses? Great, bring them. Even coarsely geo-aligned? We have a loss for that.

100

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

To train these systems, you need many examples of images with geo-located pose annotations. But the ground truth of existing datasets is unreliable and often imprecise. At the scale required for training, better ground truth is difficult to obtain.

100

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

Need an accurate camera pose, but all you have is an image and a coarse location, e.g. from GPS? No problem, just localize against OpenStreetMap. Methods unlocking that capability are called Neural Map Matchers which had poor accuracy so far. But we found the problem.

110

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 Eric Brachmann

Vladimir Yugay @vyuga3d.bsky.social · 16/07/2026

It's been several weeks since I started my research internship at @nianticspatial.bsky.social in London. Excited to finally work with @ericbrachmann.bsky.social and the team!

111

Eric Brachmann @ericbrachmann.bsky.social · 30/06/2026

Someone should create a meta-dataset and get 1 trillion citations. ... ... ... <jumps up, rushes to key board>

050

Eric Brachmann @ericbrachmann.bsky.social · 11/06/2026

Maybe we are just not ones skilled in the art.

100

Eric Brachmann @ericbrachmann.bsky.social · 09/06/2026

ACTUALLY, that very same asteroid was already discovered 1990 by the hobby astronomer, Jürgen S.

3775

Eric Brachmann @ericbrachmann.bsky.social · 28/05/2026

Oh, I know the character falling asleep all to well. In our case not randomly choose but fixed. Also cooncides with the character often choosing the movie.

220

Eric Brachmann @ericbrachmann.bsky.social · 28/05/2026

I think I am ready to be replaced by AI. My whole mentoring can be summarized by the following expert system: if True: print("Visualize everything.")

1110

Eric Brachmann @ericbrachmann.bsky.social · 28/05/2026

Better than our version: "What you wanna watch?" - "Dunno" - <3 hours of research> - "It's late, lets call it a night"

130

Eric Brachmann @ericbrachmann.bsky.social · 26/05/2026

Based on these latest KDE, what is the probability, per decade, that "at least one day in May" has the current temperature. :D

000

Eric Brachmann @ericbrachmann.bsky.social · 26/05/2026

Check you assumptions. If you are talking about 0.x% chances, you should also verify that the temperature distribution is Gaussian. :)

130

Eric Brachmann @ericbrachmann.bsky.social · 21/05/2026

I am 30 years younger than you all, still I have grown up with the disturbing visuals of the Schism video.

140

Eric Brachmann @ericbrachmann.bsky.social · 20/05/2026

I have had 70 co-authors so far. I took some time to wire all these names up with their websites on my personal page. (The BOP papers have been a real pain. 😅) Fun to browse through what all these people have been up to since we collaborated.

170

Reposted by Eric Brachmann

#CVPR2026 @cvprconference.bsky.social · 18/05/2026

We are grateful to all of the 17,491 reviewers who helped make #CVPR2026 possible. We are especially pleased to recognize the following Outstanding Reviewers, whose high-quality reviews (as judged by their Area Chairs) placed them among the top 5% of reviewers. 1/2

1258

Eric Brachmann @ericbrachmann.bsky.social · 18/05/2026

No email, a workmate found it. Indeed, Diane as well. Awesome :)

040

Reposted by Eric Brachmann

Christian Wolf @chriswolfvision.bsky.social · 24/04/2026

In my team we have two open positions on AI for robotics and in particular on manipulation, for junior and senior researchers. With us, connect fundamental research questions with real operational problems and contribute to emerging AI-driven robotics services. careers.werecruit.io/en/naver-lab...

23711

Eric Brachmann @ericbrachmann.bsky.social · 20/04/2026

It should be removed. Reviewer forms should focus on the essential content: summary, strength, weaknesses, conclusion. I usually put a summary of my review, like "interesting idea, lacks comparison to recent competitors", as a reminder for the AC which particular review this was in their stack.

150

Reposted by Eric Brachmann

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

Eric Brachmann @ericbrachmann.bsky.social · 16/04/2026

We were pretty mad at the reviewer but they turned out to be reasonable. I understand the frustration that classical methods are often overlooked. Here, it was a genuine oversight. I did not know the E5+1 solver. Including it made the paper stronger. Full discussion: openreview.net/forum?id=rmD...

openreview.net

A Scene is Worth a Thousand Features: Feed-Forward Camera...

Visually localizing an image, i.e., estimating its camera pose, requires building a scene representation that serves as a visual map. The representation we choose has direct consequences towards...

050

Eric Brachmann @ericbrachmann.bsky.social · 16/04/2026

Most multi-view methods use many (all) tokens from few images. So you have to sub-sample images to keep it computationally feasible. The motivation for FastForward is to take few tokens from many images. So you have a lean model that still utilizes many view points.

110

Eric Brachmann @ericbrachmann.bsky.social · 16/04/2026

We did not try that, but I remember seeing such a paper built on top of VGGT. If I remember correctly, it was not exactly a compact scene representation in terms of memory footprint. Definitively interesting, but expensive to cache if you wanted to use it in practise.

110

Eric Brachmann @ericbrachmann.bsky.social · 16/04/2026

FastForward has been accepted to #ICLR2026 and will be presented there next week. If you go, say hi to @axelbarroso.bsky.social. If not, don't fret. Axel has prepared a short video for you: www.youtube.com/watch?v=r0Qk...

youtube.com

[ICLR 2026] Feed-Forward Camera Localization from a Collection of Image Features

YouTube video by Niantic Spatial

260

Eric Brachmann @ericbrachmann.bsky.social · 15/04/2026

Could be a coincidence. Technically, the papers are quite different but have overlap in the motivation which is reflected in the title.

130