David Nordström

PhD Student in Deep Learning

I like Deep Learning and Computer Vision. Make GPUs go brrr.

Supervised by Prof. Fredrik Kahl and Dr. Georg Bökman.

Download Resume

News

MuM accepted at CVPR 2026

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

Feb, 2026

Pre-prints out now for MuM and RoMa v2

Released our work on Multi-View Masked Image Modeling for 3D Vision and a new dense feature matcher, RoMa v2.

Nov, 2025

ICML 2025 Spotlight

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

May, 2025

Publications

Selected

LoMa: Local Feature Matching Revisited

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

Pre-print

LoMa revisits local feature matching from a data-driven perspective, combining large and diverse data mixtures, modern training recipes, and scaled compute. We also introduce HardMatch, a new benchmark of 1000 challenging image pairs. LoMa outperforms ALIKED+LightGlue by +18.6 mAA on HardMatch and +29.5 mAA on WxBS.

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

David Nordström, Johan Edstedt, Fredrik Kahl, Georg Bökman

CVPR 2026

MuM is a feature encoder tailored for 3D vision. We extend the MAE objective to arbitrarily many frames and show that when scaling this we can beat DINOv3 and CroCo v2 on matching, feedforward reconstruction, and relative pose estimation.

Flopping for FLOPs: Leveraging equivariance for computational efficiency

Georg Bökman, David Nordström, Fredrik Kahl

ICML 2025 Spotlight Paper

We show that building flopping-equivariance into modern vision architectures reduces the number of FLOPs and increases performance.

Other

Who Handles Orientation? Investigating Invariance in Feature Matching

David Nordström, Johan Edstedt, Fredrik Kahl, Georg Bökman

CVPRW 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

Pre-print