Skimming through the API docs, it seems to be a very small subset of OpenCV at the moment.

However, a lot of vision is how you handle the details (both for speed and for accuracy/performance). Without using it extensively, it's hard to say if this is better.

OpenCV has at least a long history and being battle-tested going for it (despite its warts).