Using Computer Vision for Counting Things

Project 2023-24


Project Title: Using Computer Vision for Counting Things


Professor: Subhransu Maji


Lab/Research Group: Computer Vision Lab

Subhransu Maji co-directs the computer vision lab where he and his team of students aim to make fundamental contributions towards building AI systems with rich visual reasoning capabilities. The research focuses on architectures for visual recognition tasks, as well as techniques to improve their robustness, efficiency, generalization and interpretability. The team also engages in interdisciplinary research. In collaboration with ecologists we aim to analyze bird migration from radar imagery; with astronomers to uncover scientific insights from images of galaxies; as well as domain experts to develop applications in graphics, health care, material design, and sustainability.

Project Description:

Using computer vision and statistical estimation to solve counting in large image collections. 

Counting in large image collections (e.g., how many cars passed a traffic sign, or how many animals live in a region) is challenging, and many modern applications use computer vision to detect and count objects in massive image collections. However, when the detection task is very difficult the counts may be inaccurate even with significant investments in training data and model development. The project will build on our ongoing work called DISCOUNT (Perez et al., arXiv 2023) on combining computer vision and human screening to obtain accurate counts. The students will learn about how computer vision can be used to classify images based on modern deep learning approaches, and evaluate statistical estimation approaches for estimating counts for some applications motivated by Ecology and other domains.

Learning Objectives:

Skills needed: