Well, that's the trick. (Serial procrastinator here. Hello!) You can't think of a large project as a large project. Otherwise you'll never get it done. It's impossible by definition, because a larger project is a composite of a number of smaller tasks. So sometimes you end up having to trick yourself with "fix a typo" or "change this class name" or w/e. It has to be small and stupid enough that the activation cost is as low as possible.
Easier said than done, of course. Did I mention I'm procrastinating?