Mihai Dusmanu
Member Of Technical Staff at Anthropic
Zurich, Zurich, Switzerland
Summary
🤩
Rockstar🎓
Top SchoolMihai Dusmanu is a research-driven software engineer with a PhD from ETH Zürich and a decade of experience specializing in computer vision and deep learning for visual localization and mapping. He has held research and engineering roles at Microsoft and Anthropic, translating cutting-edge feature detection, description, and matching research into production-ready components. His open-source contributions include significant work on Hierarchical-Localization and pycolmap, adding support for SOSNet/DoG descriptors, AdaLAM matching, and Python bindings for COLMAP’s pose and transformation tools. Mihai’s work uniquely bridges academia and industry, tackling practical cloud deployment challenges such as privacy and map migration alongside algorithmic robustness. He combines low-level geometric insight with scalable engineering—optimizing memory use, edge-case handling, and RANSAC exposure—to make research usable in real systems. Based in Zurich, he continues to push localization pipelines toward reliable, privacy-aware cloud integration.
11 years of coding experience
3 years of employment as a software developer
Master's degree Mathematics Vision Learning, Master's degree Mathematics Vision Learning at ENS Paris-SaclayHigh School Diploma, High School Diploma at Tudor Vianu National College of Computer ScienceCPGE Mathematics Physics Computer Science, CPGE Mathematics Physics Computer Science at Centre International de ValbonneMaster's degree Computer Science, Master's degree Computer Science at Ecole normale supérieureDoctor of Philosophy - PhD Computer Science, Doctor of Philosophy - PhD Computer Science at ETH Zürich
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (14)
algorithms10
pytorch10
computer-vision10
eigen10
deep-learning10
python10
linear-algebra10
3d-reconstruction10
image-retrieval9
cprogramming-language9
computer-engineering9
c-language9
structure-from-motion8
cuda7
Programming languages (6)
Github contributions (5)
colmap/pycolmap
Jan 2020 - Nov 2022
(Deprecated) Python bindings for COLMAP
Role in this project:
Back-end Developer
Back-end DeveloperContributions:36 reviews, 60 commits, 20 PRs in 2 years 10 months
Contributions summary:Mihai significantly contributed to the `pycolmap` project by implementing Python bindings for core COLMAP functionality. Their work focused on integrating essential matrix, fundamental matrix and absolute pose estimation algorithms, enabling users to leverage COLMAP's features within Python. The user also added bindings for image-to-world transformations and VLFeat SIFT features. Furthermore, the user refactored the existing code by adding inlier masks and exposing RANSAC parameters, and updating camera dictionary fields.
colmappython-bindings
cvg/Hierarchical-Localization
Nov 2021 - Sep 2022
Visual localization made easy with hloc
Role in this project:ML Engineer
ML EngineerContributions:4 reviews, 8 commits, 9 PRs in 9 months
Contributions summary:Mihai primarily contributed to the feature extraction and matching components within the hierarchical-localization framework. Their work included adding support for DoG and SOSNet descriptors, modifying the existing nearest neighbor matching algorithm, and implementing AdaLAM matching. They also addressed issues related to handling edge cases, such as when no keypoints were detected, and batched SOSNet forward passes to optimize memory usage.
localizationdeep-learningvisual-localizationfeature-matchingimage-retrieval
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.