Philipp Lindenberger

The submissions to our @CVPR workshop OpenSUN3D on Open-World 3D Scene Understanding and Representations are open! 🏔️🤖☀️ ➡️ The deadline for the proceedings track is on the 5th of March!🏃‍♀️ 🌍: opensun3d.github.io 📝: openreview.net/group?id=thecv…

RaCo: Ranking and Covariance for Practical Learned Keypoints Abhiram Shenoi @PhilippCSE @pesarlin @mapo1 tl;dr: ALIKED arch + DaD RL training with full 360deg rotaug with detector+covariance heads, separate ranker (sorter) model . No IMC eval #3DV2026 openreview.net/forum?id=BWtdg…

We propose to tackle this problem by fusing two popular research branches: direct image geo-localization via classification, and cross-view retrieval against aerial imagery. We show that these tasks not only synergize, but eliminate fundamental limitations of the other approach.

At #NeurIPS2025, we're presenting our work on Scaling Image Geo-Localization to Continent Level. Website: scaling-geoloc.github.io Paper: arxiv.org/pdf/2510.26795… If you are at the conference, say 👋: 📍 Poster #4812, Dec 3 (Wed), 4:30–7:30 PM PST

@ducha_aiki No worries, thank you for sharing our work :)

Sorry, @PhilippCSE my twitter search fails you every single time :(

Some benchmark of the modern local features with LightGlue on #IMC2021 dataset. 0) Use ALIKED features. 1) ALIKED 2k is on par with other features 8k 2) DISK overfits to buildings 3) DoG detector benefits from 8k the most

@ducha_aiki Thank you for this great blog post! We also observed this phenomenon, and my best guess is that the spatial distribution of matched points and noisy pose estimation perturb the results. However, we found that non-linear pose refinement (e.g. with PoseLib) reduced this error.

@PhilippCSE do you have any idea (except possible overfitting), why match recall/precision could be not correlated with pose accuracy?

New post in my WxBS blog: Experiments with LightGlue: geometry representation and initialization. - Does it matter how you provide additional geometry info (scale, orientation for SIFT) to LightGlue? - Can we tune existing LightGlue for new features? ducha-aiki.github.io/wide-baseline-…

📢Join us tomorrow Friday morning at our #ICCV2023 poster S-6 to learn more about LightGlue! 💥We have released the training code at github.com/cvg/glue-facto… ➡️You can now train LightGlue with SIFT, ALIKED, or your own local features, on your own dataset ⏩More models coming soon!

@TobiasRobotics @condaforge @pesarlin @ducha_aiki @prefix_dev Very exciting to see pixsfm on @condaforge, thank you for this great contribution!

Excited to share that the popular Pixel-Perfect Structure-from-Motion #pixsfm is now on @condaforge. Installation is now as easy as "mamba install pixsfm" on Linux&MacOS. Note: #hloc and #colmap can be installed the same way. @pesarlin @ducha_aiki @prefix_dev #computervision 1/2

You liked SuperGlue? You'll love ⚡️LightGlue⚡️, our new deep network for light-speed image matching! ➡️Faster, stronger, easier to train than SuperGlue ➡️Code: github.com/cvg/LightGlue ➡️Paper: arxiv.org/pdf/2306.13643… Fantastic work by @PhilippCSE for #ICCV2023, with @mapo1 1/

Best Student Paper @ICCV_2021 Pixel-Perfect Structure-from-Motion with Featuremetric Refinement Philipp Lindenberger (ETH Zurich), Paul-Edouard Sarlin (ETH Zurich), Viktor Larsson, Marc Pollefeys [Session 5 A/B]

🚨 This week at #ICCV2021, check out "Pixel-Perfect Structure-from-Motion with Featuremetric Refinement" w/ @PhilippCSE V. Larsson @mapo1 ➡️ oral & *best student paper award* Website: psarlin.com/pixsfm Paper: arxiv.org/abs/2108.08291 Video: youtu.be/2HuCMuraFk0 thread ⬇️