Computer Vision Research Scientist
Exploiting photometric information in images using particular setups such as Photometric Stereo or Multi-view Helmholtz Stereo. Information about the surface orientation can be easily recovered and full and dense 3D surfaces can be reconstructed with fine geometric details.
Performing gradient descent with deformable meshes for variational problems in computer vision. Typical applications include vision problems dealing with visibility such as multi-view shape reconstruction from images, such as multi-view stereo, multi-view shape from shading or multi-view photometric stereo.
Segmenting data defined on a surface, such as color, texture, curvature or reflectance into several regions of similar attributes. This is seen as a variational segmentation problem on the manifold that is solved using a total variation based algorithm via convex relaxation.
Recovering the shape of surfaces from multi-view calibrated images. The method is based on mesh optimization, and fully account for visibility changes, in particular via the gradient of the reprojection error.
Recovering both the shape and reflectance parameters of a 3D object in a single framework.
Calibrating a 2D free-hand ultrasound probe tracked by a magnetic tracker for 3D volume generation. Blood vessels detected in the ultrasound volume can be registered to MRI image using Model-to-Image registration.