I'm disappointed by how boring the two-camera solution in the 7 Plus is. I was really hoping they were going show something interesting, like fusion of the two sensors through computational photography into one, with better low-light performance and less noise through downsampling. Or take +1 & -1 exposures simultaneously for a greater dynamic range.
Instead we got a button to switch from one lens to the other, for a 2x optical zoom. That's it. 99% of the time that second sensor and lens will do absolutely zilch. The computational bokeh appears to only use the 56mm lens for the actual photo, and the other lens to capture a depth map to compute the fake bokeh. This is hardly any better than the existing fake bokeh solutions shown many times before on Androids, Nokias and in iOS apps over the years that do the same, except it's a bit faster. The whole thing feels like just another checkbox feature, with very little actual value.
The two photos they showed to demonstrate the depth of field effect looked absolutely terrible. The edge of the mask was extremely blurry. It looked like something you could create in a photo editing app in a few seconds by just drawing over the background with a blur brush.
That is pretty much exactly what's happening, minus the manual brushing.
A depth map and face recognition is used to locate the foreground subject and a heavy blur filter is applied to the background layer. For the same reason we don't have software today that does perfect (or even halfway decent) automatic masking of faces, the edges become a blurry mess. You don't get any of the beautiful out-of-focus point of light scattering, because the lenses are incapable of capturing it.
I'm sure they tried to do something fancier with computational photography but probably ran out of time for the release. I'm confident they are going to deliver this in the typical S update next year or even as an OTA.
Instead we got a button to switch from one lens to the other, for a 2x optical zoom. That's it. 99% of the time that second sensor and lens will do absolutely zilch. The computational bokeh appears to only use the 56mm lens for the actual photo, and the other lens to capture a depth map to compute the fake bokeh. This is hardly any better than the existing fake bokeh solutions shown many times before on Androids, Nokias and in iOS apps over the years that do the same, except it's a bit faster. The whole thing feels like just another checkbox feature, with very little actual value.