The performance of the Image Understanding Architecture on The DARPA integrated image understanding benchmark

The author presents an overview of the DARPA integrated image understanding benchmark and of the Image Understanding Architecture (IUA). The performance of the architecture on various portions of the benchmark is discussed. A discussion of lessons that have been learned as a result of the experience, both in terms of the design of the architecture and of the benchmark, is included. With regard to the IUA itself, the results indicate outstanding performance on lower-level tasks, especially with the 8-b intensity data. Floating-point performance, while good, could be improved with additional hardware support. In terms of the benchmark, it is found that solving even a task as simple as this can present serious difficulties