Introduction to Deep Learning in Image Processing

• Machine Learning and Deep Learning
• Neural Network Anatomy
• Types of Convolutions
• Keras Workflow

Basic Image Processing and Computer Vision

• Pixels and Images
• Coordinate System
• Channels
• OpenCV
• Channel Ordering
• Blur and Sharpen kernels

Hands-on Lab:

• Learn basic Image Processing using OpenCV
• Learn to apply different filter kernels on images for blur generation or basic edge detection

Supervised Neural Networks and Regularization

• Underfitting
• Overfitting
• Reducing the networks size
• Weight Regularization: L1, L2, Elastic
• Dropout
• Batch Normalization

Hands-on Lab: Implement your first basic neural network, learn how to benchmark it and learn how to avoid overfitting on a Computer Vision classification task

Convolutional Neural Networks

• Convolutional Layers
• Depthwise Convolutions
• Building Convolutional Neural Networks in Keras
• 1×1 Convolutions
• Data Augmentation

Hands-on Lab: Improve your previous neural network by adding Convolutional Layers, benchmark them and compare them with the Fully Connected ones

Common Convolutional Neural Networks Architectures

• ImageNet
• AlexNet
• VGGNet
• ResNet
• MobileNet

Hands-on Lab: Learn how to use already state of the art models from the Keras Hub.

Reusing Convolutional Neural Networks

• Object Localization
• Object Segmentation
• Reusing VGG
• Fine-tuning

Real World Machine Learning

• Tensorboard
• Deploying Deep Learning Models
• Choosing the algorithm

Learn how to fine parameter tune your already trained Convolutional Neural Network to fit your task

Explainable AI

• Visualizing intermediate activations
• Visualizing convnet
• Visualizing heatmaps

Unsupervised Generative Models for Image Processing

• Autoencoders for Images
• Deblurring
• Image generation

Hands-on Lab:

• Generate a new image similar to the ones from the dataset by using a random seed
• Face generation techniques

Real World Machine Learning

• Tensorboard
• Deploying Deep Learning Models
• Choosing the algorithm