Framework for checking algorithm accuracy

Does anyone know of a framework for checking accuracy of algorithms? Essentially something along the lines of unit tests, but for cases when we know that there will be failures, but would like to get the accuracy, and perhaps an optional list of failed cases.

I intend to use it for detection and classification algorithms (say detecting humans in images) to easily check performance of different algorithms against test suite. I could of course write my own tool, or maybe coerce a unit testing framework to the task, but perhaps there is a dedicated framework available I'm not aware of?

I'm not limiting myself to any specific language for now.