Pixel Recurrent Neural Networks

Explore the cutting-edge field of image generation through an in-depth 30-minute lecture on Pixel Recurrent Neural Networks. Delve into the three dominant approaches in the field, examine typical architectures, and understand the crucial role of kernel masks. Review Recurrent Neural Networks (RNNs) and their application to image generation, with a focus on Long Short-Term Memory (LSTM) equations. Analyze input-to-state components, state-to-state components, and learn how to combine state components effectively. Conclude by examining various model architectures, gaining valuable insights into this innovative area of machine learning and computer vision.

Syllabus

Intro
Outline
Three image generation approaches are dominating the field
Typical Architecture
Kernel mask
Masks
RNN Review
RNN for Image Generation
LSTM Equations
Input-to-State Component
Finished State-to-State Component
Combine State Components
Model Architectures