Artificial Intelligence in Computer Vision: From Convolution to Transformers

Course objective:

The goal of the course is to familiarize participants with the practical implementation of more advanced deep neural network models such as convolutional and transformer networks in the Tensorflow Keras environment. Participants of the course will gain theoretical knowledge about the individual layers that build such models and learn high-level tools that allow for their easy use. The course will also discuss and present in practice the transfer learning methodology, which allows for the adaptation of ready-made architectures to your own data sets. Participants will use the acquired knowledge in practice by training several convolutional and transformer models for image classification.

Who is the target audience?

The training is intended for beginners who have never had contact with convolutional networks and transformers. Basic knowledge of model construction using the Keras library and the correct selection of model parameters (see the course "Introduction to Keras") is needed. Advanced programming skills are not required.

Program:

1. Convolutional networks - a reminder
2. Pre-trained architectures (VGG, ResNet, EfficientNet, DenseNet, ConvNext etc.)
3. Transfer learning methodology
4. Fine-tuning
5. Vision Transformers (ViT)
6. Network interpretability (GradCAM++, ScoreCAM, SaliencyMaps, Occlusion, LIME)
7. Attention of transformer models
8. Uncertainty analysis (Monte Carlo Dropout Uncertainty)

How to sign up?

1. To register for a course, you need an account in our system. To set it up, click here. This account allows you to use the offer of the entire training platform, including courses conducted by CDSI and other AGH units, which can be accessed using the tabs at the top of the page. If you already have an account, you can skip this step.
2. After creating an account, sign up for the course. Acceptance is based on the order in which applications are submitted.
3. You will receive automatic confirmation of your application from our system, which will include the transfer details.
4. Send payment confirmation to the e-mail address kursy-cdsi@agh.edu.pl.