At a Glance
- University of Massachusetts Amherst computer scientists developed DISCount, an AI framework that rapidly detects damaged buildings in crisis zones and accurately estimates bird flock sizes.
- The framework blends AI’s speed and data-crunching power with human analysis to deliver reliable estimates from large image collections, earning recognition for its social impact.
- DISCount emerged from collaborations with the Red Cross to count damaged buildings and with ornithologists to estimate bird flock sizes. It aims to provide more reliable automated tools.
- By combining existing AI computer vision models with human analysis, DISCount significantly reduces labeling costs and surpasses alternative screening approaches, offering confidence intervals for informed judgments.
- The framework has been successfully demonstrated in applications such as counting birds in radar imagery for climate change studies and assessing damaged buildings in satellite imagery for disaster response, showcasing its potential in various fields.
Computer scientists at the University of Massachusetts Amherst have developed an AI framework called DISCount, which addresses the challenges of quickly detecting damaged buildings in crisis zones and accurately estimating the size of bird flocks. This innovative framework combines the speed and data-crunching power of artificial intelligence with the reliability of human analysis to deliver accurate estimates from extensive collections of images. The research, published in the Proceedings of the AAAI Conference on Artificial Intelligence, has been recognized by the association with an award for the best paper on AI for social impact.
The framework, DISCount, emerged from the team’s collaboration with the American Red Cross to develop a computer vision tool for accurately counting damaged buildings in crises and their work with ornithologists interested in estimating the size of bird flocks using weather radar data. The team realized that the standard computer vision models were not accurate enough for these tasks and sought to build automated tools that could provide a higher degree of reliability while being accessible to non-AI experts.
DISCount fundamentally rethinks the typical approaches to solving counting problems by combining the strengths of human analysis and AI. It leverages existing AI computer vision models to analyze large data sets and determine smaller ones for human researchers to examine. This human-in-the-loop approach allows for more accurate and reliable counts, significantly reducing labeling costs and surpassing alternative screening approaches.
The framework has been demonstrated in two applications: counting birds in radar imagery to understand responses to climate change and counting damaged buildings in satellite imagery to assess damage in regions affected by natural disasters. DISCount not only reduces labeling costs but also provides confidence intervals, enabling researchers to make informed judgments about the accuracy of their estimates.
References
- Perez, G., Maji, S., & Sheldon, D. (2024). DISCount: Counting in Large Image Collections with Detector-Based Importance Sampling. Proceedings of the AAAI Conference on Artificial Intelligence, 38(20), 22294–22302. https://doi.org/10.1609/aaai.v38i20.30235
- University of Massachusetts Amherst. (2024, April 11). New computer vision tool can count damaged buildings in crisis zones and accurately estimate bird flock sizes. Tech Xplore. https://techxplore.com/news/2024-04-vision-tool-crisis-zones-accurately.html
