The project will initially examine 30 non-violent and violent protest campaigns since 2010 in countries such as Egypt, Bosnia-Herzegovina, Burkina Faso, Guatemala, Ukraine, Burundi, Thailand, Venezuela, and Pakistan. For each case, the research team scrapes photos from Twitter, Facebook, and Google Images that were taken during the protest events. Using a machine learning computer vision tool, these photos will be analyzed for the gender composition of the crowd, interactions between groups, incidence of violence, and protest activities. By using images spanning the course of a campaign, this process will provide evidence as to who participated and how throughout the protests, as well as changes in movement structure and trends in protests tactics. Changes in crowd participation over time will allow us to analyze how these variables are related to the likelihood of movement success or failure, or the likelihood that a movement will shift from non-violent to violent tactics. The dataset of our findings and the tool used to analyze them will advance our ability to understand how gender dynamics shape episodes of contention and violence.