As much as I like Fowler, he's missing an important option in his description: don't try to do object relational mapping generically. Consider that you have business model objects. In the UI, you create a "view" of that data. Of course the UI view is not related in structure to the business model object. That would be absurd. We need to use collections of model objects and present the data in a flexible. Just as absurd would be to build a framework for automatically mapping the data in the model layer to the UI view.
Now consider that your DB is just like your UI. It's a view of your model objects. Just in the same way, there is no reason why your model objects should be related structurally to your DB representation of the data. And in exactly the same way there is no particular reason why you should have a way to automatically map to and from your DB layer to your model objects.
In many cases, it will be less complex to build a bespoke OR mapping rather than try to find a system that will do it generically.
Now consider that your DB is just like your UI. It's a view of your model objects. Just in the same way, there is no reason why your model objects should be related structurally to your DB representation of the data. And in exactly the same way there is no particular reason why you should have a way to automatically map to and from your DB layer to your model objects.
In many cases, it will be less complex to build a bespoke OR mapping rather than try to find a system that will do it generically.