YOLOD – Improving neural nets for detecting packed cars in satellite imagery

Neural network research moves very quickly and it can be hard to keep up. In our work at Radiant Solutions, we use a variety of machine learning modalities. My team works with neural networks for segmentation, detection, and other imagery processing tasks, but our most common application is object detection. Detection is basically the task …

Introducing the Metis/DigitalGlobe Data Challenge

Last year, I completed the Metis Data Science Bootcamp in New York. I finished hiking the entire Appalachian Trail at the end of 2015, then spent 6 grueling months looking for a job. Once I accumulated a few hundred rejection letters, and it seemed like I wasn’t any closer to a job, I realized my skills …

Measuring Performance for Object Detectors – Part 2

This post is the second and final installment in my series about measuring the accuracy of models. In my last post, I talked about how to choose a method to score models. Now it’s time to use that method to score models against each other. Let’s start with a brief summary of the goal. We …

Creating Synthetic Clouds in Python

Tiny Changes Can Fool AI There has been much discussion more recently (and some not so recently) on how minute changes to images can fool the smartest neural nets. Sharif et al. showed how to fool a neural net into classifying a Reese Witherspoon photo as Russell Crowe by adding a groovy pair of technicolor …

Measuring Performance for Object Detectors – Part 1

This is the first part of a 2-part series of posts about measuring the accuracy of detector models. As we develop more models, it’s becoming important to have a standard score that tells us how well each model performed on a problem so we can choose the best one for each application. In order to …