People with visual impairments (PVI) face poorer quality of life, mental and physical health, as well as low socioeconomic status. Few low-cost tools exist that assist PVI effectively. The rising potential of computer vision provides a unique opportunity to assist people with low vision in real-time with minimal cost and at a large scale. Here we analyze a large-scale dataset from a smartphone application that automatically detects objects using computer vision and also reads text. To our knowledge, this is the first study to examine the use of computer vision to assist blind people in real-life scenarios on a large scale. Characterizing the needs of people with visual impairments is a critical step for developing and improving the next generation of assistive technologies as well as novel artificial vision solutions. We find that VI users used the application for reading, finding objects, and exploring their environment. Top searched objects included person, seat, and door, possibly to help the user with wayfinding. Most users were from developing countries, and many used the applications for extended durations (over 20 minutes). These initial findings are a first step in understanding the use of computer vision for PVI at a large scale.# project-poster