Lewis Fishgold, Author at Azavea - Beyond Dots on a Map

Ground truth imperfectly overlaying machine learning predictions.

Using Noisy Labels to Train Deep Learning Models on Satellite Imagery

By Lewis Fishgold on August 5th, 2019

How do noisy labels affect the accuracy of a deep learning model? We added different amounts of noise to the SpaceNet Vegas buildings dataset and trained some models to find out.

Raster Vision applied to the city of Georgetown, Guyana.

Software Engineering

Raster Vision 0.9 Release Candidate

By Lewis Fishgold on April 15th, 2019

This release of Raster Vision includes bug fixes, an easier setup, improved performance, and the ability to train models off of labels in OSM.

Software Engineering

Predicting Land Use in the Amazon using Deep Learning

By Lewis Fishgold and Annie Zhao on January 3rd, 2018

Learn about our methods and results leveraging our RasterVision deep learning project to predict land use in satellite images of the Amazon.

Lewis Fishgold speaking about RasterVision at our Technical Staff Meeting

Software Engineering

Video: Using Convolutional Neural Networks to Automatically Analyze Aerial and Satellite Imagery

By Lewis Fishgold on September 6th, 2017

In this recording of our most recent Technical Staff Meeting, we walk through our team’s work on Raster Vision, a...

Software Engineering

Deep Learning for Semantic Segmentation of Aerial Imagery

By Lewis Fishgold and Rob Emanuele on May 30th, 2017

Update (10/2018): Raster Vision has evolved significantly since this was first published, and the experiment configurations that are referenced are...