Sorbonne Université

Master 2 IMA - TADI

UE VISION

Schedule 2025-2026: Tuesday afternoons

ROOMS: See https://cal.ufr-info-p6.jussieu.fr/master/

Lecturers: Dominique Béréziat (SU / LIP6), Antoine Manzanera (ENSTA / U2IS), Pascal Monasse (ENPC / Imagine)


Objectives
 
The objective of the COMPUTER VISION course is to present the techniques of geometric (3d) and kinematic (movement) analysis of a scene by a vision system, from the mathematical / physical / biological principles, to the algorithmic implementation.

The purpose here is not to recognise the objects or to identify a scene, but to geometrically structure a scene and recover the motion of the objects and the camera.

However, geometry and movement are fundamental cues for image understanding.

Displacement analysis is related to 3d reconstruction by the common problem of feature matching between different views of the same object.

Evaluation Mode

This course is long and continuously evaluated,
which requires a certain amount of diligence:
  • 2 or 3 practical works/projects on Geometric 3d.
  • 1 oral presentation on Biology & Co-design.
  • 1 practical work on Motion Estimation.
  • 1 practical work on Motion Detection & Tracking.

  TIME 
SESSION DESCRIPTION
LECTURER RESOURCES
Sep, Tue. 23 1:45pm
  • Introduction / 3d & Movement - Biological Perception: role of motion in biological vision, local vs global motion perception, Gestalt, Optical flow and Time-before-contact, Binocular and Monocular perception of 3d
  • 3d & Movement - Co-design approaches: Active 3d RGB-d cameras - Time-of-Flight, structured light; Passive monocular depth sensors - Plenoptic cameras, Depth from (de)focus, Coded aperture; Focal plane processing - Event-based cameras, programmable retinas;
Antoine Manzanera
Sep, Tue. 30 12:45pm
  • 3d Geometry from images: Projective geometry, Camera matrix, panorama construction
Pascal Monasse
Oct, Tue. 7 12:45pm
  • 3d Geometry from images: Essential and Fundamental Matrices, their calculation, RANSAC algorithm
Pascal Monasse
Oct, Tue. 14 12:45pm
  • 3d Geometry from images: Local Estimation of Disparity, correlation
Pascal Monasse
Oct, Tue. 21 12:45pm
  • 3d Geometry from images: Global Estimation of Disparity, maximal flow and graph-cut for the calculation of disparity maps
Pascal Monasse
Nov, Tue. 4 12:45pm
  • 3d Geometry from images: Multi-view Stereo
Pascal Monasse
Nov, Tue. 18 1:45pm
  • Oral presentations on Biological vision & Co-design: Please email the subject of your presentation to Antoine Manzanera before November, 3
Antoine Manzanera
Dec, Tue. 2 1:45pm
  • 3d & Movement - Machine learning based approaches: Supervised and unsupervised methods for optical flow prediction, Supervised methods for single-view depth map prediction, Unsupervised methods for odometry and depth map prediction
Antoine Manzanera
Dec, Tue. 9 1:45pm
  • Motion detection and tracking: Background estimation and subtraction methods, Markovian and morphologic regularisation, Distribution based methods, Hough based methods, Mean-Shift, Predictive Filtering
Antoine Manzanera
Dec, Tue. 16
1:45pm
  • Motion estimation: Frequence based methods, block matching, variational methods for local and global optical flow estimation, regularisation methods
Dominique Béréziat
Jan, Tue. 6
1:45pm
  • Motion estimation #2
Dominique Béréziat
Jan, Tue. 13 1:45pm
  • Practical work on Motion estimation #1
Dominique Béréziat
Jan, Tue. 20 1:45pm
  • Practical work on Motion estimation #2
Dominique Béréziat
Jan, Tue. 27 1:45pm
  • Practical work on Motion detection & tracking
Antoine Manzanera