Explore an innovative approach to inverting multi-channel 2D remote sensing data into 3D volumetric models using deep learning techniques. Learn about a novel convolutional architecture that dramatically reduces processing time from months to milliseconds. Discover fascinating analytical methodologies, including an embedding strategy using eigenvector decomposition, 2D to 3D information conversion in latent space, and an extensive investigation of various model architectures. Gain insights into the use case, technical details, and a demonstration using physically realistic synthetic data. Understand the challenges faced, successful model architectures, and the process of determining optimal feature transforms. Engage in a discussion about potential improvements in accuracy and training speed, and explore applications in computer vision, digital signal processing, and volumetric space manipulation.
Inversion of 2D Remote Sensing Data to 3D Volumetric Models Using Deep Dimensionality Exchange
Overview
Syllabus
Introduction
Project Description
Why Care
Physical Inversion
Past Work
Problems
Model Architecture
Synthetic Data
Transfer Learning
Convergence criterion
Whats left
Example dashboards
Taught by
Open Data Science
