START: A tool for Rapidly Generating Image-Truth and Evaluating ATR Performance

S. Ralph, J. Irvine, M. Stevens, and M. Sorrason

Proceedings of WACV, Breckenridge, CO (January, 2005)

ATR algorithm evaluation requires thorough testing, using diverse data sets and several performance metrics. This is a challenging effort due to the very time consuming generation of image-truth data, heretofore generally created by a human operator on a frame-by-frame basis. It is also very tempting when confronted with limited resources to implement only one type of ATR scoring metric. The Scoring Truthing and Registration Toolkit (START) is being developed to address these two important aspects of ATR evaluation: an application that greatly expedites the generation of high-fidelity image truth (the Truthing Compoment), and a Scoring Component that computes and reports a set of scoring metrics on the output from a candidate ATR algorithm.

START architecture

Request a Copy