Technical skills
Languages, frameworks and infrastructure I work with daily.
Python
Expert6 years · primary language
PyTorch
Expert6 years · deep learning
MATLAB
Advanced3 years · signal processing
Git / GitHub
Expert5 years · version control
Cloud & MLOps
Advanced2 years · scalable workflows
Focus area
Multimodal foundation models for computer vision
Scaling transformers that read images, geometry and physics together.
Commit heatmap
Past 365 days, aggregated from GitHub and self-hosted Gitea.
Tip: swipe horizontally to view the full year.
Note: We switched to a self-hosted Gitea instance in December 2025, so activity from the previous GitLab instance cannot be displayed here. Most research code lives privately on our department's Gitea — selected projects are mirrored publicly on GitHub below.
Pinned repositories
Public mirrors of selected research code on GitHub.
AstroFetch
ML-ready planetary science data for PyTorch, starting with the Moon.
multimodal-lunar-swirls
Code for "Probing Lunar Swirls" — a multimodal masked autoencoder over 12 remote-sensing modalities across 56 swirl sites.
lunar-technosignatures
Code for the VISAPP 2025 paper — deep-learning anomaly detection of the Apollo landing sites in LRO NAC imagery.
More on github.com/TechnicToms — the remaining research repositories are hosted privately on our department's Gitea instance and available on request.
The Moon's many faces
A single unified transformer for multimodal lunar reconstruction
We formulate reflectance-parameter estimation and image-based 3D reconstruction of lunar images as a single multimodal learning problem. One transformer learns shared representations across grayscale images, digital elevation models, surface normals and albedo maps — translating freely from any input modality to any target.
Predicting DEMs and albedo maps from a single grayscale image simultaneously solves surface reconstruction and disentangles photometric parameters from height — a foundation model that learns physically plausible relations across all four modalities.
Details
- Type
- Journal Article
- Field
- Planetary Science & ML
- Role
- First author
- DOI
- 10.1016/j.isprsjprs.2026.04.008
Sander, Tenthoff, Wohlfarth & Wöhler — TU Dortmund, Image Analysis Group.
1.34B
Tokens processed by our transformer
4
Modalities, one unified model
29.7M
Trainable parameters
Q1
Journal, first-author publication
Selected work
More work in preparation — see the papers page for the complete list, talks and posters.