AllWrite: the first Internet-based platform for on-demand benchmarking handwriting recognition systems
Résumé
This paper describes the AllWrite platform which innovates in the field of handwriting recognition system benchmarking. The AllWrite architecture provides the handwriting recognition community with a common benchmarking infrastructure designed (1) to host the training/test data sets together with their related ground truth, (2) to compute the metrics used for the objective performance evaluation, (3) to support any recognition system via an open and easy-to-use API, and (4) to render the evaluation outputs. Taking advantage of the RIMES project experience, AllWrite goes one step beyond, offering several advantages. It ensures an on-line, on-demand independent and fair evaluation. It solves the over-training problem. It makes automatic the system run and the result analysis phase hence saving a large amount of human resources. If AllWrite is dedicated to researchers, it is also well-suited for the on-line evaluation of any industrial system, the later remaining agnostic to and independent of the whole AllWrite software complexity