Hacker News Reader

The article/advert misses this information out; is this depth estimation a model in the OS somewhere, or part of the vision api, coremedia api (where the stereo->depth frames are), or just inside the camera app and not exposed to developers?

View on HN

Author here. In the article, we mentioned it's a neural network, and the results are accessible to developers. We left out further details because this isn't an implementation guide.

Could you give me a hint here which of their API's it's using? There's a few "depth" frame interfaces in iOS for various devices and googling which apply to which devices rarely yields much. (I wish the apple docs would hint as to which devices apply to which APIs)

In all probability, there will be no true depth sensing capability in SE, thus developer would not be able to access the api.

The portrait mode is likely to be implemented with monocular depth estimation using deep learning.

Of course it's a depth-from-mono model, it says so in the article. Apple's "true depth" is an estimate from two cameras, that's not actually real depth either but it's exposed to developers.

Also in the article "We’re happy that this depth data — both the human matte and the newfangled machine-learned depth map — was exposed to us developers, though."

Reply on news.ycombinator.com