An Open Architecture for End-to-End Document Analysis Benchmarking

Abstract

In this paper, we present a fully operational, scalable and open architecture allowing end-to-end document analysis benchmarking without needing to develop the whole pipeline. By decomposing the analysis process into coarse-grained tasks, and by building upon community provided state-of-the art algorithms, our architecture allows any combination of elementary document analysis algorithms, regardless their running system environment, programming language or data structures. Its flexible structure makes it straightforward to plug in new algorithms, compare them to other algorithms, and observe the effects on end-to-end tasks without need to install, compile or otherwise interact with any other software than one's own.