Post-doctoral Researcher (M/F) in Computer Vision (H/F)
Role details
Job location
Tech stack
Job description
This position is part of the CNRS "OPEN" valorization project named DOPAMIn. It aims to transfer state-of-the-art 3D reconstruction technology (RNb-NeuS), derived from academic research, into the open-source industrial software AliceVision/Meshroom, in collaboration with the University of Zurich. Mission The researcher will integrate, optimize, and stabilize the RNb-NeuS neural 3D reconstruction method within the AliceVision software architecture, enabling the creation of high-precision digital twins.
Activities
Activity 1: Integration of Photometric Stereo in Meshroom
- Implement processing nodes for normal field and intrinsic color estimation.
- Integrate deep learning-based methods (such as SDM-UniPS or Uni-MS-PS).
Activity 2: Implementation of the RNb-NeuS Module
- Develop the neural reconstruction node using photometric data as input.
- Ensure interfacing with the AliceVision library.
Activity 3: Critical Code Optimization (C++/CUDA)
- Adapt the code to drastically reduce computation times (target: < 1h).
- Replace proprietary dependencies (InstantNGP) with a flexible base (such as SuperNormal).
Activity 4: Reliability and Robustness
- Develop mechanisms for outlier rejection.
- Document the code for the open-source community.
Requirements
Knowledge:
- In-depth expertise in Computer Vision and Photogrammetry.
- Mastery of state-of-the-art Neural Rendering (NeRF, NeuS, SDF).
- Knowledge of Photometric Stereo methods.
Operational Skills:
- Advanced programming in C++ and Python.
- Mandatory mastery of GPU programming (CUDA) for optimization.
- Experience with Deep Learning frameworks (PyTorch).
- Knowledge of the AliceVision architecture is a major asset.
Soft Skills:
- Autonomy and scientific rigor.
- Ability to work collaboratively (mixed academic/industrial context).
- Ability to meet short development deadlines (project mode).
