On the 360 Video Professionals Facebook group there was one report which said it looked good, another was not so enthused:
https://www.facebook.com/groups/360.vid ... 413226072/
I want to see some close up content -- and with a moving camera -- it is all very well having a static camera and having most things in the scene more than a couple of meters away. The Google claim implies that this is an automatic algorithmic process with no need of human tweaking. We will see.
One thing the demonstration view of the workflow for the workshop scene shows is just how imprecise Gopros are in their sensor alignment. Successive slices from the cameras have frames up or down 10s of pixels from adjacent ones -- with big exposure differences from slice to slice. It makes me think that is is still a terrible workaround hack with all those memory cards to wrangle, upload etc etc
Also, the obvious thing -- this makes cylinders, not spheres -- why not modded Gopros with 180 fisheye coverage -- because Gopro wouldnt like it!
The basic concept of extracting depth maps to improve frame interpolation is a an old one, behind all but the most basic retimers -- eg, Twixtor, Ocula etc -- Google's primary innovations will be how well they stitch the slices and if they really can provide position tracking in a stereo 360 video in a limited planar circular zone.
