David Nordström

David Nordström

PhD Student in Deep Learning

I do
Deep Learning 
Supervised by
Fredrik Kahl
and
Georg Bökman
.

News

Pre-print released for Poincar3, emerging multi-view geometry from self-distillation without labels. Project page.

3 papers accepted at ECCV 2026: LoMa (Oral), Octic ViTs, and RoMa v2 (Oral). Also, RoMa-Ω accepted at the SFM-DL workshop.

MuM accepted at CVPR 2026. Our paper on Multi-View Masked Image Modeling for 3D Vision.

Pre-prints released for MuM and RoMa v2.

"Flopping for FLOPs" accepted as Spotlight Paper at ICML 2025.

Selected

LoMa: Local Feature Matching Revisited teaser figure

LoMa: Local Feature Matching Revisited

David Nordström*, Johan Edstedt*, Georg Bökman, Jonathan Astermark, Anders Heyden, Viktor Larsson, Mårten Wadenbäck, Michael Felsberg, Fredrik Kahl

ECCV 2026(Oral)

Revisits local feature matching with large data mixtures and scaled compute. +18.6 mAA on HardMatch over ALIKED+LightGlue.

MuM: Multi-View Masked Image Modeling for 3D Vision teaser figure

MuM: Multi-View Masked Image Modeling for 3D Vision

David Nordström, Johan Edstedt, Fredrik Kahl, Georg Bökman

CVPR 2026

Feature encoder for 3D vision via multi-frame MAE. Beats DINOv3 and CroCo v2 on matching, reconstruction, and pose estimation.

Octic Vision Transformers: Quicker ViTs Through Equivariance teaser figure

Octic Vision Transformers: Quicker ViTs Through Equivariance

David Nordström, Johan Edstedt, Fredrik Kahl, Georg Bökman

ECCV 2026

Builds 8-fold symmetry into Vision Transformers, reducing compute while improving performance through equivariance.

Flopping for FLOPs: Leveraging Equivariance for Computational Efficiency teaser figure

Flopping for FLOPs: Leveraging Equivariance for Computational Efficiency

Georg Bökman, David Nordström, Fredrik Kahl

ICML 2025(Spotlight)

Flopping-equivariance in modern vision architectures reduces FLOPs and improves performance — a free lunch from symmetry.

* Equal contribution.

Other

Poincar3: Emergent Multi-View Geometry Through Self-Distillation

David Nordström, Thibaut Loiseau, Vincent Lepetit, Michael Felsberg, Guillaume Bourmaud, Fredrik Kahl

Pre-print, 2026

RoMa v2: Harder Better Faster Denser Feature Matching

Johan Edstedt, David Nordström, Yushan Zhang, Georg Bökman, Jonathan Astermark, Viktor Larsson, Anders Heyden, Fredrik Kahl, Mårten Wadenbäck, Michael Felsberg

ECCV 2026(Oral)

RoMa-Ω: What Feed-Forward 3D Models Know About Image Matching

David Nordström, Xinyue Zhang, Thibaut Loiseau, Vincent Lepetit, Fredrik Kahl

ECCV 2026(Workshop)

Who Handles Orientation? Investigating Invariance in Feature Matching

David Nordström, Johan Edstedt, Fredrik Kahl, Georg Bökman

CVPR 2026(Workshop)

Supervision

Students I have advised, and the work that came out of it.

Does VGGT Know What It Does Not Know?

Adityan Thirumani Srinivasan, David Nordström

Pre-print, 2026

Advised Adityan Thirumani Srinivasan on his first research project, carried out during his final year of high school. The paper shows that a linear probe on VGGT's frozen features predicts pose failure far better than the model's own confidence.

Impact

Citations

130

+40 last 3 months

h-index

5

+1 last 3 months

GitHub stars

1,566

+175 last 3 months

Papers

8

+2 last 6 months

Cumulative, by month · updated Nov 2026

050100150200Mar '25Jun '25Sep '25Dec '25Mar '26Jun '26Sep '26Dec '26May 2025 — ICML 2025 SpotlightNov 2025 — RoMa v2 + MuM pre-printsFeb 2026 — MuM @ CVPR 2026Apr 2026 — LoMa pre-printJun 2026 — 3× ECCV 2026 (2 Orals)Sep 2026 — Poincar3 pre-print

◆ marks a milestone. Citations come from Semantic Scholar, so they run lower than Google Scholar's count.

Awards & Recognition

Conference Distinctions

Oral at ECCV 2026 (2 papers) and Spotlight at ICML 2025.

2025–2026

Outstanding Reviewer, NeurIPS 2026

Selected for outstanding reviewing contributions.

2026

SSBA Best Paper Award

Swedish Society for Automated Image Analysis.

2026

Open Source

davnords — bash

Just need the matcher? uv add lomatch works as a drop-in replacement for LightGlue in any SfM or visual localization pipeline.

Education

Chalmers

PhD in Geometric Deep Learning and 3D Computer Vision

Chalmers University of Technology

Ongoing

Research focus on equivariant neural networks, efficient deep learning architectures and 3D computer vision.

Chalmers

M.Sc. in Engineering Mathematics

Chalmers University of Technology

International Experience

Exchange Semester at UC Berkeley

Exchange Semester at Seoul National University

Chalmers

B.Sc. in Industrial Engineering

Chalmers University of Technology

5.0/5.0 GPA
University of Gothenburg

B.Sc. in Economics

University of Gothenburg

Completed in parallel with Industrial Engineering degree

Teaching

Computer Vision

EEN020, Chalmers University of Technology

Fall 2025

Deep Machine Learning

SSY340, Chalmers University of Technology

Fall 2025

Talks

Learning to Understand 3D in the Age of Feed-forward Reconstruction

Invited talk at École des Ponts ParisTech (ENPC), Paris, while visiting Vincent Lepetit. October 2026.

Slides

Learning to Understand 3D

Invited talk at the Workshop on Geometry, Learning and 3D Understanding — a pre-ECCV mini-workshop at the Centre for Mathematical Sciences, Lund University. September 2026.

Slides

Feedforward 3D Reconstruction

Guest lecture in the Computer Vision course at Chalmers University of Technology.

Video

Acknowledgements

My research is a result of a great collaboration with
Georg Bökman
and
Johan Edstedt
, and my main supervisor
Fredrik Kahl
, they have ben instrumental to all of my works. I am also grateful for the nice funding by the Wallenberg AI, Autonomous Systems and Software Program (WASP) which has given me many great benefits.

Get in Touch

I'm always open to research collaborations and discussions.

Feel free to reach out via email for any inquiries or opportunities.