There’s more and more 360 imagery, but if it’s made using a Gopro Max, while a great camera, horrible when it comes not having metadata during it’s timelapse photo mode that would allow for horizon leveling the images.
Either you (Mapillary) make a tool for us to do that somehow, (that tries to auto guess it and we just need to verify or something). Or you manage to convince Gopro to add that metadata to the images. I can’t imagine it being THAT hard for them to do … but so far they haven’t done it…
Human heads aren’t perfect camera gyro’s so def during a bike ride you’ll get images like this:
I have a 360 camera which adds metadata. The problem is still there though, as you cannot trust the values provided by the camera accelerometers if you mount it on a helmet, unless you’re stationary and make a point of holding your head still.
Hugin can try to auto guess, and I think that it supports scripting. However, without manual intervention I mostly get bad results because of lack of vertical features.
Until it becomes easier to auto-level time-lapse photos from the GoPro Max I highly recommend filming or using time warp, even though the resolution is of lower quality. The benefit of filming is more photos with a smaller interval. This usually makes up for the lower quality. The user experience using Mapillary heavily suffers if the camera direction is all over the place.
@asturksever sorry for the ping. But do you guys actually have some kind of documentation on all the EXIF tags that are read by Mapillary?
I’m borrowing the Max again this month and have a whole afternoon of footage that isn’t level (camera arm must have move. all my other days were fine, just this one afternoon).
And I’d like to manually fix the horizon level
But do you guys actually have some kind of documentation on all the EXIF tags that are read by Mapillary?
You can read here about minimum Exif requirements (and frankly the only Exif tags processed on the backend):
However, currently no amount of Exif modification will help you level the horizon (for Mapillary). For this to happen the web viewer would need to consider GPano:PosePitchDegrees and GPano:PoseRollDegrees, which it currently does not.
Note also that leveling the horizon only serves an aesthetic purpose, especially when viewing an image. Reconstruction on the backend does not care about the horizon being level because by design it automatically computes the position, pitch, roll, and yaw of the camera with respect to other images, so that results are almost always (as long as there is enough visual overlap between images and enough control points match) horizon correct.
And I’d like to manually fix the horizon level
Your only viable option for now to get an aesthetically pleasing effect in the Mapillary web viewer is to re‑project your images.
Your only viable option for now to get an aesthetically pleasing effect in the Mapillary web viewer is to re‑project your images.
Yeah, not going to do that because of quality loss
Sure it’s aesthetic and backend doesn’t care. But we do see usage of municipalities and stuff from our imagery, so for them aesthetics does matter. (it’s weird looking at crooked imagery)
Hi @Thibaultmol - this is something that can be done automatically with computer vision by Mapillary - we will add this to our backlog, unfortunately due to competing priorities I’m guessing it won’t be available in the short term. Thank you for the feature request, it makes a lot of sense, and we hope to make this available in 2024.
