Lightnews — Scholar-powered news

Eric Brachmann

@ericbrachmann.bsky.social

2K followers 540 following 360 posts

Niantic Spatial, Research. Throws machine learning at traditional computer vision pipelines to see what sticks. Differentiates the non-differentiable. 📍Europe 🔗 http://ebrach.github.io

ebrach.github.io

Posts Media Videos Starter Packs

Pinned

Eric Brachmann @ericbrachmann.bsky.social · 13d

⏩ FastForward ⏩ A new model for efficient visual relocalization. Accurate camera poses without building structured 3D maps.

nianticspatial.github.io/fastforward/

Work by @axelbarroso.bsky.social, Tommaso Cavallari and Victor Adrian Prisacariu.

1 3 27

Eric Brachmann @ericbrachmann.bsky.social · 3h

I did not realize the first author is actually here! 🙈 @wenjingbian.bsky.social

Eric Brachmann @ericbrachmann.bsky.social · 3h

@ericzzj.bsky.social is doing a better job than me finding people here on BlueSky. Here is Leonard: @roym899.bsky.social

Eric Brachmann @ericbrachmann.bsky.social · 4h

🌟 Scene Coordinate Reconstruction Priors at #ICCV2025 🌟

Can we learn what a successful reconstruction looks like, and use this knowledge when reconstructing new scenes?

Explainer: youtu.be/RkV6U5xYc20
Project Page: nianticspatial.github.io/scr-priors

Inspiring work by Wenjing Bian et al.!

1 9

Reposted by Eric Brachmann

Walter Scheirer @wjscheirer.bsky.social · 15h

The #ICCV2025 main conference open access proceedings is up:

openaccess.thecvf.com/ICCV2025

Workshop papers will be posted shortly. Aloha!

8 20

Eric Brachmann @ericbrachmann.bsky.social · 1d

I agree. Leonard is a Viz-ard.

Eric Brachmann @ericbrachmann.bsky.social · 1d

🔥 ACE-G, the next evolutionary step of ACE at #ICCV2025 🔥

We disentangle coordinate regression and latent map representation which lets us pre-train the regressor to generalize from mapping data to difficult query images.

Page: nianticspatial.github.io/ace-g/

Stellar work by Leonard Bruns et al.!

2 5 22

Eric Brachmann @ericbrachmann.bsky.social · 2d

No.

patrick star and spongebob squarepants are standing next to each other in a dark room .

ALT: patrick star and spongebob squarepants are standing next to each other in a dark room .

media.tenor.com

Eric Brachmann @ericbrachmann.bsky.social · 5d

I hope I said something sensible. 😅 Great work!

Eric Brachmann @ericbrachmann.bsky.social · 5d

When that work includes formulas.

In unrelated news: Looking forward to our #ICCV2025 RANSAC tutorial.

a man wearing headphones is sweating while sitting in an airplane .

ALT: a man wearing headphones is sweating while sitting in an airplane .

media.tenor.com

Eric Brachmann @ericbrachmann.bsky.social · 5d

When you present other peoples work with them sitting in the audience.

a man with sweat running down his face

ALT: a man with sweat running down his face

media.tenor.com

1 6

Eric Brachmann @ericbrachmann.bsky.social · 5d

Congratulations! That thesis is a hell of a package.

Eric Brachmann @ericbrachmann.bsky.social · 7d

Just from todays feed: bsky.app/profile/kmyi...

Kwang Moo Yi @kmyid.bsky.social · 7d

Yugay and Nguyen et al., “Visual Odometry with Transformers”

Instead of point maps, you can also directly output poses. This used to be much less accurate, but now it's the opposite. Simple architecture that directly predicts camera embeddings, which then regress rot and trans.

1 1

Eric Brachmann @ericbrachmann.bsky.social · 7d

I think it is a great time to have such a tutorial again. As we see competitive RANSAC-free approaches arise, it is worth looking back - and looking forward.

Dmytro Mishkin @ducha-aiki.bsky.social · 7d

For those going to @iccv.bsky.social, welcome to our RANSAC tutorial on October 2025 with
- Daniel Barath
- @ericbrachmann.bsky.social
- Viktor Larsson
- Jiri Matas
- and me
danini.github.io/ransac-2025-...
#ICCV2025

1 2 7

Reposted by Eric Brachmann

Dmytro Mishkin @ducha-aiki.bsky.social · 7d

1 6 20

Eric Brachmann @ericbrachmann.bsky.social · 7d

Whenever I do a voice over, I realize that I sound like Arnold Schwarzenegger when I say "coordinate". Unfortunately, in my line of work, I have to say "coordinate" a lot...

Eric Brachmann @ericbrachmann.bsky.social · 11d

I thought you just turned 34?

Eric Brachmann @ericbrachmann.bsky.social · 12d

Trendy.

Vision and Graphics Trends @si-cv-graphics.bsky.social · 12d

𝗔 𝗦𝗰𝗲𝗻𝗲 𝗶𝘀 𝗪𝗼𝗿𝘁𝗵 𝗮 𝗧𝗵𝗼𝘂𝘀𝗮𝗻𝗱 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀: 𝗙𝗲𝗲𝗱-𝗙𝗼𝗿𝘄𝗮𝗿𝗱 𝗖𝗮𝗺𝗲𝗿𝗮 𝗟𝗼𝗰𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗳𝗿𝗼𝗺 𝗮 𝗖𝗼𝗹𝗹𝗲𝗰𝘁𝗶𝗼𝗻 𝗼𝗳 𝗜𝗺𝗮𝗴𝗲 𝗙𝗲𝗮𝘁𝘂𝗿𝗲𝘀
Axel Barroso-Laguna, Tommaso Cavallari, Victor Adrian Prisacariu, Eric Brachmann
arxiv.org/abs/2510.00978
Trending on www.scholar-inbox.com

Eric Brachmann @ericbrachmann.bsky.social · 12d

Great to have Kwang on Bluesky! (And quite flattering that this is his first post 😳)

Kwang Moo Yi @kmyid.bsky.social · 12d

Barroso-Laguna et al., "A Scene is Worth a Thousand Features: Feed-Forward Camera Localization from a Collection of Image Features"

When contexting your feed-forward 3D point-map estimator, don't use full image pairs -- just randomly subsample! -> fast compute, more images.

Eric Brachmann @ericbrachmann.bsky.social · 12d

The model estimates image poses relative to images of which you already know the poses. It solves a sub-problem of what COLMAP generally is used for. FastForward is a visual relocalization model, not a full structure-from-motion model.

Eric Brachmann @ericbrachmann.bsky.social · 12d

*whisper* Only one way to know...

Eric Brachmann @ericbrachmann.bsky.social · 12d

Oh no, today of all days...

1 5

Reposted by Eric Brachmann

Zhenjun Zhao @ericzzj.bsky.social · 13d

A Scene is Worth a Thousand Features: Feed-Forward Camera Localization from a Collection of Image Features

@axelbarroso.bsky.social, Tommaso Cavallari, Victor Adrian Prisacariu, @ericbrachmann.bsky.social

arxiv.org/abs/2510.00978

1 1 3

Eric Brachmann @ericbrachmann.bsky.social · 13d

FastForward was trained on challenging frame sets. Therefore, it is able to solve difficult relocalization cases, like opposing shots in the video below. We only have mapping images of the front of the sign. Still FF relocalizes well even behind the sign.

Eric Brachmann @ericbrachmann.bsky.social · 13d

FastForward uses 20 retrieved images by default. But no need to compute attention over 21 images. To keep the model lean, we sub-sample the feature tokens of mapping images.

Each token is combined with its ray embedding. Hence, FastForward operates on samples of visual tokens in scene space.

1 1

Eric Brachmann @ericbrachmann.bsky.social · 13d

Building visual maps via reconstruction (SfM, SCR) can take time. Instead:

⏩ Get poses for your mapping images, e.g. via SLAM in real time.
⏩ Build a retrieval index in seconds.
⏩ FastForward predicts the relative pose between a query image and all retrieved images in one forward pass.