Using computer vision to gain health insights from social media

Abstract

In this thesis, I developed machine learning methods that would support tools for gaining health insights from social media images. On the one hand, I have helped create a dataset for food image segmentation and a segmentation network for this task, a task that is very relevant for understanding the nutritional content of food from images. I also explored the flip side of the issue, and helped design an interface that users can use to explore food recipes in machine learning-driven ways. On the other hand, I helped create a dataset and model for classifying types of disasters given images of natural or manmade disasters.