I made a test walk the other morning…
My 360cam is mounted on a monopod, tied to the alu frame of a backpack. The monopod is then raised to avoid as much of me as possible in the nadir area.
Any comment on this sequence?
Should I try to raise the monopod any further?
Yes, I know about the light. The LG360 needs more light to avoid blurry pics, and this sequence was taken really early in the morning. 
Looks good, cam is raised high enough to see all sides clear. Btw, your 360 cam pictures show much less stitch artefacts than mine cam does…
How many seconds interval ? The pictures are still connected.
Did you walk slowly ?
I suppose it is not possible on a bike to have connected pictures, certainly with such monotonous background.
Really nice pictures. Your setup is OK
Wow, this is insanely good! And I get this feeling of this “cloud ceiling” that you can actually touch by just raising up your arm 
Thank you everyone! 
I use the default setting in the iOS app: Distance: 5m and interval 2s.
I’m a pretty fast walker when the ground surface is flat.
I think it looks like a great setup. The camera is pretty level and I think you are getting maximum image quality.
Some of the images are not level when you look to the sides, e.g. Mapillary
That is because the camera is not perfectly level. It saves metadata to allow for correction of this afterwards and there is an old bug on https://github.com/mapillary/mapillary_issues/issues/2009 that Mapillary should do that for us. If they implement that, we can get perfectly level pictures without any stabilization hardware at all. Please go in and comment on it
The question is whether it’s because of the cam or the field, or a combination.
I know that the field has a slight slope towards the last third, then it’s going up again.
In this case it may be due to the field, but in other cases like Mapillary it is definately the cam that was not level. Which is not easy when biking.
Oh, I see…
And it can’t be easy either, to level up the cam as you lean forward over the handle bar, cause when you shift position the cam will be tilted the other way… Especially if you go up- and downhill all the time.
Would it be better to set it up on a monopod behind your saddle?
I don’t attach the camera to the bike, because it would give a lot of problems with vibrations. This would cause a lot of blur and probably break the camera in the end. It would not be a perfect solution either. But because the camera records the information nessesary to correct any tilt in any direction, the perfect solution actually exists. You can hold the camera in any way and get a perfectly level image in the end.
One can discuss where the operation should be performed: By me, in Mapillarys backend or in Mapillarys viewer. If Mapillary does it, they can always do it better later, but if I do it, there is no option to correct it. For this reason I think it is best if Mapillary does it.
I can’t see any problem (bar the loss of quality due to the processing and recompression ) as long as while rotating the image before upload, you also update the tilt information accordingly. Mapillary could still apply a different rotation, if necessary, to an image which has already been levelled before upload.
What I’d like to see on Mapillary site,is an option to manually adjust the angles of a choosen picture, based on a visual feedback, just like you can adjust direction and position of shooting. I shoot while walking, and probably shake the camera too much. As a result, the recorded tilt direction is often slightly inaccurate, After rotating the picture to zero the recorded tilt angles, it still looks skewed. Adjusting it would require user interaction, unless Mapillary could do some magic guess based on image content interpretation?
That would be a good option yes. The question is what to do if any information in the corners are missing? Zooming in, or filling out with something?
There is a similiar feature in Photoshop Elements, for leveling up family photos. Maybe have a look at how they solve it?
For a standard picture, I’d say just show an inclined rectangle when fully zoomed out. If you don’t like it, you can zoom in in the viewer. Speaking of a 360 cam, that’s not a problem anyway, since there are no corners, you can rotate any way you want.