I will no reach SOTM but there is an interesting topic:
A review of Mapillary-generated map data and how accuracy compares across devices
This experimental study shows that positioning accuracy is highly related to the GPS accuracy of the capture device, and in general, a large part of the final positional error can be attributed to this. Mapillary’s 3D reconstruction is able to mitigate some of these effects.
Just use RTK GNSS.
We are developing a free and collaborative RTK network in France and it is running well. Public regional networks are also available in Italy (for STOM?). I just used the one in north-west (SPIN3 GNSS). For instance : Mapillary
Indeed, I’m not sure that the Mapillary app under Android is fully taking advantage of it, especially along the track (time stamp accuracy?).
Overall, about survivability of Mapillary, one can observe that contributor experience is poor. It must be a central point in Mapillary, as provider of raw data. For one year, time to time, I’m loosing data. Failed sequences after upload from android app. One can not correct pointing direct of pictures. One can not delete a single picture in a sequence (like pictures recorded while entering a covered parking and positioned hundreds of meters away). One can not retrieve our own pictures not blurred and at full size.