All students must validate 30 ECTS per semester.We offer two speciality tracks:

Track 1: Vision and Applications
The first track focuses on various applications of computer vision: biomedical applications, people detection, object tracking, computational photography.

Track 2: Vision and devices
The second track focuses on devices to capture images (intelligent sensors, medical imaging systems) and to visualise and interact with them (augmented reality)

 

 

 

Course name: Deep Learning in Computer Vision Credits: 3

  • Class type: lecture/practical
  • Hours per week: 4h
  • Type of the exam: practical works + final exam
  • Prerequisites (if exist): basic image processing algorithms, programming in C++

Content

The course is devoted to the methods of supervised learning applied in computer vision and in particular to the innovative group of methods, such as Deep Learning. During the course various aspects of Deep learning will be covered. The fundamentals in neural networks such as MLP will be presented. Then, the focus will be on the variety of convolutional neural networks and associated problems, such design of architectures, data preparation, optimization methods, fine tuning/domain adaptation and fusion with Deep architectures. The temporality aspects will also be considered for various computer vision, CBIR and CVIR applications. In practical work students will acquire skills for the use of optimal network configurations accordingly to the available OpenSource frameworks

Required Readings

V. Vapnik, “The Nature of Statistical Learning Theory”, series Statistics for Engineering and Information Sciences,  2010.

Recommended Readings

I. Goodfellow, Y. Bengio, "Deep Learning", series Adaptive Computation and Machine Learning, MIT Press, 2016.
J. Benois-Pineau, P. Le Callet, "Visual Content Indexing and Retrieval With Psycho-visual Models", Eds, Springer, 2017.
R. Girshick, J. Donahue, T. Darrell, J. Malik, "Rich feature hierarchies for accurate object detection and semantic segmentation", arXiv:1311.2524
G. Csurka, Domain Adaptation in Computer Vision Applications. Advances in Computer Vision and Pattern Recognition, Springer 2017,
B. Chu, V. Madhavan, O. Beijbom, J. Hoffman, T. Darrell, “Best Practices for Fine-Tuning Visual Classifiers to New Domains”. ECCV Workshops (3) 2016: 435-442

  • Lecturer (name, position, degree): Dr. Jenny Benois-Pineau, Full Professor
  • Additional lecturers, if exist(name, position, degree): assistants (PhD students...)

The European Credit Transfer and Accumulation System (ECTS) is a student-centred system based on the student workload required to achieve the objectives of a programme of study. Its aim is to facilitate the recognition of study periods undertaken abroad by mobile students through the transfer credits. The ECTS is based on the principle that 60 credits are equivalent to the workload of full-time student during one academic year.