SEED-GRANT: Understanding Image-based Big Data using Human Computation
This proposal seeks to enhance machine understanding of big image and video data sets by including human computational units as an element inside larger computational systems. Cameras generate by far the most data in the world, and even small organizations can easily deploy cameras to monitor social and environmental problems.
Thus, this project has four goals: characterize and benchmark human computation co-processors (HPUS) platforms, propose instructions and algorithms, implement instructions and algorithms, and evaluate work. Researchers seek to build an automated test suite of small tasks that can be run on a wide set of HPU platforms, providing a snapshot of the accuracy, latency, and reliability of each platform. Researchers need to evaluate their work, which will be done using applications already under development that have previously suffered from insufficient robustness of existing computer vision libraries. Lastly, researchers will quantify performance against existing solutions.