How to use 360 camseras in RealityScan Desktop
https://dev.epicgames.com/community/learning/tutorials/o9ke/using-360-cameras-in-realityscan
Can pano2views tool remove people automatically?
Does this require or work best with any particular version of RealityScan?
This is a solution to a problem Iām currently struggling with, so Iām excited to see it!
Will there be a way to use 3D180 - 2x square images from a stereoscopic ārigā camera derived from 180deg fisheye for processing? 360 is nice but requires the operator / gimbal etc to be masked out which isnāt part of this tool..? (and perhaps would be better to process as individual 180fisheyes anyway since the rig introduces positioning error due to offset, plus equirect stitching introduces other image errors, saying it would probably be more beneficial to support fisheye lens models full stop)
Iāll be honest in that Iām disappointed in how low effort this feels from RS.
Not only is the ā360 camera supportā not actually a native feature in RealityScan, but the generated XMPās donāt actually create camera rig constraints which lock the positions of the cropped views to one another.
Meaning, you STILL end up with badly aligned cropped views within a single 360 capture.
1 Like
No but itās certainly something we will be looking into
This is fantastic feedback. Currently our SFM does not support fisheye models and itās a non-trivial task to add them. You will find with some models of camera the fisheye distortion coeficients in the exif, so you can experiment with your own undistorts, or go the open CV chess board route. I tried this and as you say the results for front back are probably slightly better than going the equirectangular route, but without fisheye in SFM this requires a bespoke solution for every camera. If you do go this route donāt make your pinholes too wide, for this reason i actually found that even going straight from fisheye our SFM much prefered (i donāt remember the precise stats butā¦) letās say 4 images in a ring of 120 FOV rather than 2 back to back 170 FOV pinholes. Also great feedback on the masking 
very interested in any other thoughts you have or features you would be interested in, I am assuming also rigging?
This is great feedback! You are absolutely right proper rigs would really help here. Please let us know if there are any other features you would like to see.
No it should work the same with most versions. It just turns the inputs into something RealityScan can process
Hi @WallaceWain, I realize fisheye model is difficult (hence mostly no software supports it beyond 120degFOV, let alone PolyFisheye model) but I would think it would well be worth the effort creating a pathway that is not splitting & pinholes . Also was wondering, my Qoocal3Ultra (which I have one modded into a 3D180 8K) also records proper IMU, gyro and GPS data, some of which exists as GPX and others can be extracted with tools like Gyroflow. (I would expect this to work well since this is a stereoscopic rig which is well suited to depth reconstruction If you can trust computer vision. )
When I import video clips into RS2 none of that is being seen / understood / extracted. Would be nice to be able to create the trajectory and extrinsics with the data that is already there, no? Would like to marry this with some LiDAR scan data