3IA PhD/Postdoc Seminar #40

Abstract:
Symmetric positive definite matrices, particularly correlation matrices, have become a common source of geometric information in neuroimaging. The challenges encountered with the conventional Euclidean metric have led to the development of alternatives based on Riemannian metrics. It turns out that correlation matrices are widely used to describe brain connectivity in anatomical and functional neuroimaging. In this short presentation, we will first explore several methods to endow the space of correlation matrices with a differentiable structure. Next, we will present recent Riemannian metrics developed for full-rank correlation matrices and demonstrate how these metrics enable us to perform statistics on non-linear spaces. Finally, we will illustrate our work with applications in the study of brain connectomes, using data obtained from rs-fMRI.

Temporally Propagated Masks and Bounding Boxes: Combining the Best of Both Worlds for Multi-Object Tracking

Abstract:
Multi-object tracking (MOT) involves identifying and consistently tracking objects across video sequences. Traditional tracking-by-detection methods, while effective, often require extensive tuning and lack generalizability. On the other hand, segmentation mask-based methods are more generic but struggle with tracking management, making them unsuitable for MOT. We propose a novel approach, McByte, which incorporates a temporally propagated segmentation mask as a strong association cue within a tracking-by-detection framework. By combining bounding box and propagated mask information, McByte enhances robustness and generalizability without per-sequence tuning. Evaluated on four benchmark datasets - DanceTrack, MOT17, SoccerNet-tracking 2022, and KITTI-tracking - McByte demonstrates performance gain in all cases examined. At the same time, it outperforms existing mask-based methods.

Solving parametric systems of equations with neural network

Abstract:
We consider a parametric system of  equations F(X,P)=0 where P are parameters and X the unknowns. When the P are fixed F(X) is a square system of n equations in the n unknowns in X. We will show that using classical approaches of NN the solutions predicted by the NN are usually extremely bad. But we will show that by using structured training sets and hybridized multi-layer perceptrons it is possible to build a very fast solver that provide exact solutions as soon as the set of P is bounded.

Open discussion about the two contributions