Hello Mapillary users! This thread is a place to post feedback about the Mapillary plugin for JOSM. Add a comment here with any issues you have, any feature requests, discussions of existing useful features, and bug reports or crash details. I will lead the Mapillary team discussions about JOSM topics so we can directly see your feedback and use it to plan the next generation of tools.
I remember mapillary once offered a point cloud view on their website. Might it be possible to project the point cloud to 2D to get a map layer? Similar to the features layer, but with much more raw geometry points. That could help a lot with mapping where current orthophoto or satellite imagery is not (yet) available, or where trees cover the top view. Those points would either have object color or be colored according to machine learning classification. My hope would be that one can see structures from this new map layer, but of course I am not sure if this actually can work.
For a while I have seen a strange behaviour of the JOSM Mapillary plugin: some, but not all traces have erratic deviations if compared with the same trace on the Mapillary.com site, in iD, in RapidiD.
Using JOSM 17919 and Mapillary plugin v2.0.o-alpha.36* on Windows 10.
Looks like this particular case is a problematic 3d scene reconstruction which caused the computed locations to go a bit haywire. Not a bug but just a bad outcome. They should not be so drastically different so this sequence is getting a deeper look, but many things could cause this (camera angle issues, corrupted data, other things, can’t be sure until we dive in).
I believe the JOSM plugin actually defaults to the computer locations, and in the settings menu for the plugin you can choose which is default and change it. Let me know if you see that working.