M2R - MSc in Informatics

Graphics Vision and Robotics
and  Ubiquitous Interface Systems

Computer Vision 

MoSIG M2 2020-2021 Academic Year

Zoom - Thursdays 9:45  to 12:45

Professors: James L. Crowley and Edmond Boyer

Teaching Assistants: Yantao Wang and Nachwa Aboubakr

These class notes can be found at http://crowley-coutaz.fr/jlc/Courses/2020/GVR.VO/GVR-VO.html
The original planned Class schedule.  Here is a pointer the the ADE class reservation system
Programming Teams

Source Files for lecture Notes

Thursday 1 Oct 2020    Lesson 1:  Theory: Performance Evaluation for Recognition and Detection 

Course Introduction:  James Crowley
       •  Course Organisation
Computer Vision Theory: (Nachwa Aboubakr)
  • Pattern Recognition and Machine Learning
  • Performance Evaluation Evaluation Metrics
  •              Exercise Questions (exam questions from past years)
    Practical Instruction
    Jupyter Notebooks, OpenCV, and  FDDB.
             • Using OpenCV and Keras in Python with Jupyter Notebooks
             • Opening and displaying a face with the FDDB Data set
    Programming Exercise 1:  Displaying Faces from the FDDB data set
                Examples of Best Notebooks: Team2-Domps-Zhu.ipynb, Team4-Handowsk-Granzio.ipynb, Team11-Brusca-ALsaka.ipynb
    Background Reading:
    The FDDB Data Base (Jain and Learned-Miller 2010)

    8 october 2020 - ENSIMAG Partners Day - no classes

    Thursday 15 Oct 2020  Lesson 2:  Visual Perception in Man and Machine (Recorded Lecture)

     Computer Vision Theory (Recorded Lecture for Part 1)
  • Albedo and Reflectance
  • The Human Visual System
  • Vergence, Version and Fixation
  • Color Perception and Color Spaces
  •               Exercise Questions (exam questions from earlier years)
    Practical Instruction (Recorded Lecture for Part 2)
               • Sliding Window Face Detectors
              •  Programming Neural Networks in Keras
              •  Detecting Faces with a 3 Layer MLP in Keras
    Programming Exercise 2Face Detection with a Multi-Layer Percetron
    Evaluation Data for Exercise 2 (from folds 9 and 10 of FDDB)
        Examples of Best  Notebooks:
    Team12-HMedan-Zhong-Lab2.ipynb, Team2-DOMPS-ZHU-Lab2-buildDataset.ipynb, Team2-DOMPS-ZHU-Lab2-MLP.ipynb
    Background Reading:
    (Rowley and Kanade 87)

    Thursday 22 Oct 2020  Lesson 3:  Scale Space and Image Pyramids (Recorded Lecture

    Computer Vision Theory   
  • Scale Space
  • Gaussian function as a low-pass digital filter
  • Scale Invariant Gaussian  Pyramids
  • Equivariance Properties of Scale Space
  • Practical Instruction:   ( Recorded Lecture)
    Yangtao Wang's Jupyter Notebook demo
    Constructing an Image pyramid with OpenCV
               • Detecting Faces at multiple scales with a pyramid
    Programming Exercise 3Detecting Faces at multiple scales with a pyramid and a sliding window MLP (updated 25 oct) - Due on 11 Nov.
    Background Reading: 
    Face Detection with Half octave Pyramid (Ruiz 2008(Crowley-Riff 2003)

    29 Octobere 2020 Fall Vacation (Toussaint) - no classes

    Thursday 5 Nov 2020  Lesson 4:  Local Image Description with Receptive Fields   (Recorded Lecture - Theory Part)

    Computer Vision Theory:  (Recorded Lecture)
  • Gaussian Derivatives
  • Using the Gaussian to compute image derivatives
  • Steerability and Intrinsic Orientation
  • Intrinsic Scale
  • Histogram Of Oriented Gradients  Recorded Lecture for Practical Part
  • Scale Invariant Feature Transform (SIFT)
  • Practical Instruction: Revised Version of Exercise 3 - Due 12 Nov. - Presentation Slides of Revised Exercise 3   Recorded Lecture for Practical Part
              • Yangtaos's Jupiter notebook for Half-Octave Gaussian Pryamid
    Background Reading:
    SIFT Paper (Lowe  1999)
                • Fast Computation of Receptive Fields - (Crowley-Riff 2003)
                • Fast computation of Characteristic (Intrinsic) scale - (Crowley-Piater 2003)
                • Face Detection with Gaussian Derivatives (Ruiz-Crowley 2008)

    Thursday 12 Nov 2020   Lesson 5:  Attention and Cognition for Computer Vision (Recorded Lecture)

    Practical Instruction: Detecting and Tracking Faces in Video Sequences - (Recorded Lecture)
              • Bayesian Tracking process:  Predict, Detect, Update
              • Tracking with Adaptive Background Subtraction
              • Face Tracking with Skin Color Blobs

    Example of OpenCV code to display video squences  ( Zip of file for download)
    Programming Exercise 4
    Experimental Performance Evaluation for Face Tracking.

    Computer Vision Theory: Attention and Cognition for Computer Vision - (Recorded Lecture)
              • Cognitive Vision
              • Visual Concepts and Visual Attention
              • Structured Knowledge Representations
              • Situation Models.

    Background Reading:  
            Robust Face Tracking using Color - Schwerdt-Crowley 2000
            Benchmarking Face Tracking - Fischer et al 2011
            Evaluating Multiple Object Tracking Performance - Bernadin2008

    Thursday 19 November 2020  Lesson 6:   Homogeneous Coordinates and Projective Camera Models (Recorded Lecture)

     Computer Vision Theory:  (Recorded Lecture) Homogeneous Coordinates and Projective Camera Models (Recorded Lecture Part 1, Part 2)
           • Homogeneous Coordinates and Tensor Notation
            • Homogeneous Coordinate Transforms
            • Homographies
            • The Projective Camera Model
            • Camera Calibration

    Practical Instruction:   More on Bayesian Tracking:  Tips and Techniques (Recorded Lecture for Practical Part)

    Instructions and Advice for mid-term report

    Midterm Project Report  Definition and Grading Scale (50% of the final Grade)

            Report on performance evaluation for different techniques for face detection and face racking
            Report due by E-mail on 4 January 2020

    Lesson 7:   Structure from Motion (Edmond Boyer)

    Thursday 26 Nov 2020  -  9:45 to 12:45  

    Lesson 8: Structure from Motion (Edmond Boyer)

    Thursday 3  December 2020  -  9:45 to 12:45

    Lesson 9 Shape Models (Edmond Boyer)

    Thursday 10  December 2020   -  9:45 to 12:45 

    Lesson 10: Articulated Motion (Edmond Boyer)

    Thursday 17  December 2020 -  9:45 to 12:45

    Thursday 25 December 2020 Christmas holidays - no class

    Thursday 1 january 2021 - New Years holidays - no class

    Lesson 11: Articulated Motion (Edmond Boyer)

    Thursday  7 janvier 2021   -  9:45 to 12:45 

    Lesson 12: Articulated Motion (Edmond Boyer)

    Thursday  14 janvier 2021   -  9:45 to 12:45 


    Face Detection Data Sets:

    FDDB dataset: FDDB dataset contains the annotations for 5,171 faces in a set of 2,845 images.
    WIDER FACE: A face detection benchmark dataset with 32,203 images and labels for 393,703 faces with a high degree of variability in scale, pose and occlusion.
    Head Pose Data Set:  Created by Nicholas Gourier in 2004.
    MALF dataset: Face Detection in the Wild.  MALF consists of 5,250 images and 11,931 faces.
    AFW dataset: Face Detetion in the Wild. AFW dataset is built using Flickr images. It has 205 images with 473 labeled faces.
    For each face, annotations include a rectangular bounding box, 6 landmarks and the pose angles.

    Face Tracking Data Sets:

    This link gives direct direct access to the individual directories the AVDIAR: A Dataset for Audio-Visual Diarization.
    The data-set includes 27 short stereo video recordings of people walking and talking, along with hand-labeled bounding boxes. The audio files have been removed from this version.
    ( See https://team.inria.fr/perception/avdiar/  for the complete multi-modal data set and additional documentation. )

    Exercise Problem Sets / sample exam questions (Do these at home or in a group)

    Lesson 1: Performance Evaluation
    Lesson 2:
    Human Vision: Attention and Fixation
    Lesson 3:
    Pyramids, Scale Space
    Lesson 4: Gaussian Receptive Fields
    Lesson 5: Bayesian Detection and Tracking
    Lesson 6: Projective Transformations

    Past Exams

    M2R GVR 2016:    Computer Vision Exam from November 2016
    M2R GVR 2014:    Computer Vision Exam from November 2014
    M2R GVR 2012:    Computer Vision Exam from November 2012
    M2R GVR 2011:    Computer Vision Exam from November 2011
    M2R GVR 2010:    Computer Vision Exam from November 2010
    M2R GVR 2009:    Computer Vision Exam from September 2009