Explore unlearned neural networks as image priors for inverse problems in imaging through this 50-minute online lecture from Northeastern University's CS 7180 Spring 2020 class. Delve into Deep Image Prior, Deep Decoder, and Deep Geometric Prior concepts, examining their applications in super-resolution and denoising. Analyze the architectural differences, parameter considerations, and representation capabilities of these approaches. Investigate the geometric picture, smoothness locality, and over/under-parameterization effects. Gain insights from related papers on Image Adaptive GAN and Latent Convolutional Models. Access accompanying lecture notes for a comprehensive understanding of these cutting-edge techniques in artificial intelligence and computer vision.
Unlearned Neural Networks as Image Priors for Inverse Problems
Overview
Syllabus
Introduction
Neural Networks for Inverse Problems
Deep Image Prior
Geometric Picture
Super Resolution
Questions
Deep Decoder
Deep Decoder Architecture
Deep Decoder Parameters
Deep Decoder Representation
Denoise
Deep Geometric Prior
UnderParameterize Deep Decoder
OverParameterize Deep Decoder
Smoothness Locality
Image Adaptive Gann
Taught by
Paul Hand
