Baptiste Brument Thesis Defense

We are pleased to announce the PhD defense of a thesis dedicated to improving the robustness and accuracy of 3D reconstruction methods by integrating explicit models and leveraging physical (reflectance, shading, refraction) and geometric (gradients, silhouettes) properties.

Rather than proposing entirely new approaches, this research enhances existing geometric and differentiable rendering frameworks, extending their applicability to challenging scenes and materials.

The work introduces four main contributions:

1. MVS with image gradients – improving surface normal estimation and matching precision.
2. 3D reconstruction in refractive media – modeling refraction explicitly for multi-view stereo (RMVS).
3. Shading-based surface reconstruction (SfS-NeuS) – resolving ambiguities on weakly textured surfaces.
4. Photometric prior re-parametrization (RNb-NeuS) – robustly fusing normal and reflectance data.

This thesis advances the pursuit of high-fidelity geometry reconstruction in complex and previously unresolved imaging scenarios.

https://bbrument.github.io