arXiv preprint ·
Guiding Image-to-3D Generation with Test-Time Partial Observations
Guiding pretrained image-to-3D models with partial geometric observations at test time, without retraining or finetuning.
PaperComputer vision · Geometry · Optimisation
Simon Weber.
Research Associate · University of Oxford
I study geometry and optimisation at the intersection of 3D computer vision and artificial intelligence.
Research AssociateUniversity of Oxford
I am a Research Associate at the University of Oxford in the PIXL lab headed by Ronald Clark and an ELLIS member. I received my PhD in Computer Vision from the Technical University of Munich in the Computer Vision Group, headed by Daniel Cremers . I visited the Dyson Robotics Lab at Imperial College London, under the guidance of Andrew J. Davison . Before that, I studied at École normale supérieure (ENS Ulm) in Paris.
My research interests include 3D reconstruction for structure-from-motion and SLAM systems, beyond-Euclidean geometry, and semantic segmentation.
Latest news
A few recent updates- Our paper on asymmetric manifold learning has been accepted at NeurIPS 2026!
- Our paper on rethinking attention has been accepted at ICML 2026!
- I became an ELLIS member.
Older news 10 updates
- I joined the University of Oxford, PIXL lab!
- I successfully defended my PhD in computer vision.
- Visiting Andrew Davison's lab at Imperial College London for a few months!
- Our Finsler Multi-Dimensional Scaling paper has been accepted at CVPR 2025!
- Our CVPR paper Flattening the Parent Bias: Hierarchical Semantic Segmentation in the Poincaré Ball is presented at the workshop Beyond Euclidean at ECCV 2024.
- Our Power Variable Projection for Initialization-Free Large-Scale Bundle Adjustment paper has been accepted at ECCV 2024.
- Visited International Computer Vision Summer School in Sicily.
- Our Flattening the Parent Bias: Hierarchical Semantic Segmentation in the Poincaré Ball paper has been accepted at CVPR 2024.
- Our Finsler-Laplace-Beltrami Operators with Application to Shape Analysis paper has been accepted at CVPR 2024.
- Our Power Bundle Adjustment for Large-Scale 3D Reconstruction paper has been accepted at CVPR 2023.
Preprints
2026* Equal contribution
arXiv preprint ·
Initialization-Free Bundle Adjustment Revisited: A Controlled Experimental Study
A controlled study of why low optimisation error does not always yield accurate metric 3D reconstructions in initialization-free bundle adjustment.
PaperPublications
2021–2026* Equal contribution · † Project lead
NeurIPS, 2026
Harnessing Data Asymmetry in Manifold Learning
Exploring asymmetric relationships in data through manifold learning.
Paper
ICML, 2026
Functional Attention: From Pairwise Affinities to Functional Correspondences
Rethinking attention through functional correspondences beyond pairwise affinities.
Paper
PhD Dissertation, Technical University of Munich, 2025
From Krylov Spaces to Power Series: Advancing Large-Scale Bundle Adjustment with and without Initialization
Methods for large-scale bundle adjustment, with and without an initial estimate.
Dissertation
CVPR, 2025
Finsler Multi-Dimensional Scaling: Manifold Learning for Asymmetric Dimensionality Reduction and Embedding
Using Finsler geometry to embed data with asymmetric relationships.
Paper
ECCV, 2024
Power Variable Projection for Initialization-Free Large-Scale Bundle Adjustment
Large-scale bundle adjustment without requiring an initial reconstruction.
Paper
CVPR, 2024
Flattening the Parent Bias: Hierarchical Semantic Segmentation in the Poincaré Ball
Addressing parent bias in hierarchical semantic segmentation with hyperbolic geometry.
Paper
CVPR, 2024
Finsler-Laplace-Beltrami Operators with Application to Shape Analysis
Bringing Finsler geometry to differential operators for shape analysis.
Paper
CVPR, 2023
Power Bundle Adjustment for Large-Scale 3D Reconstruction
A power-series approach to bundle adjustment for large-scale 3D reconstruction.
Paper
GCPR, 2021
Multidirectional Conjugate Gradients for Scalable Bundle Adjustment
Exploring multidirectional conjugate gradients for scalable bundle adjustment.
Paper