Deep Learning for Vision (DLV)

Description

This course studies the typical architectures and applications of deep learning models to computer vision problems such as image classification, object detection, semantic segmentation and visual content generation. After recalling the bases of machine learning with neural networks and the main tools to understand modern neural architectures, we will dive into the typical methods developed to tackle computer vision problems.

Keywords

Visual representations, convolutional neural networks, classification, regression, object detection.

Prerequisites

Basic knowledge of Linear Algebra, Calculus, Probabilities, Machine Learning, Python, C++

Contents

Part 1 - Bases in deep learning (7h30, Elisa Fromont)
  • Intro ML and main computer vision (learning) problems (1h30). The slides are available here.
  • Neural Networks Basics (3h45). The slides are available here.
    • Perceptron, MLP, Backprop
  • Deep learning (2h15). The slides are available here.
    • Convolutional Neural Networks
    • Recurrent Neural Networks (LSTM, GRU)
    • AutoEncoders and Generative Models
Part 2 - Deep learning for vision (12h, Denis Coquenet)
  • Vision architectures for feature extraction (VGG, Resnet, Vision Transformer) : 3h00
  • Object Detection dedicated architectures (YOLO, RCNN) : 3h
  • Semantic segmentation architectures (FCN, U-Net, ...) : 1h30
  • Generative models for vision : 3h
    • GAN
    • Diffusion Models
  • Applications (Handwriting recognition) : 1h30
Oral presentations (1h30)

Teachers

Elisa Fromont (responsible), Denis Coquenet