/training-deep-learning-models-for-vison

Primary LanguageJupyter Notebook

Archived because the course is finished. See https://github.com/constantinpape/dl-teaching-resources for updated versions of the exercises.

Training Deep Learning Models for Vision - Compact Course

Join the chat at https://gitter.im/training-deep-learning-models-for-vison/community

A compact course on using deep learning methods for computer vision. Based on material from the EMBL deep learning course.

See the course website for details on the version of this course taught at Heidelberg University in October 2020.

Requirements

In order to follow this course you should have basic knowledge of machine learning and/or computer vision and have python programming experience. The course will be taught using Google Colab, in order to use it you need a google account. Some free space on your google drive (~ 3GB) will be benefitial Google Drive.

Content

Day 1:

  • Deep Learning for Vision
    • Basics of Machine learning, focus on supervised learning
    • Basics of Neural Networks: MLPs and SGD
    • Deep Learning Frameworks and pytorch

Practical part:

  • Data preperation for pytorch
  • Image Classification on CIFAR (Logistic Regression -> MLP)

Day 2:

  • Deep Learning for Vision:
    • Introduction to CNNs
    • More on training and data augmentations
    • Advanced architectures: ResNet etc.

Practical part:

  • Image classification with CNN on CIFAR10
  • Data Augmentation and advanced architectures on CIFAR10

Day 3:

  • Image segmentation and denoising with U-Net (in detail)

Practical part:

  • Semantic Segmentation on DSB2018
  • Denoising with noise-to-noise approach

Day 4:

  • Deep Learning for Object detection

Practical part: Work on longer exercise.

Day 5:

Practical part: Work on longer exercise.