Power users better use mapillary_tools for now because it features among other things the ‑‑duplicate_distance and ‑‑duplicate_angle options, which enable not only tuning duplicates but also image/frame density between GPX track points extracted from video. The Desktop Uploader currently lacks such and other sequence creation control features.
The Insta 360 X5 camera is a milestone in spherical distortion correction. The X5 features initial approaches to a converter that makes .inf data directly usable for Mapillary.
Insta360 X3 and the problem with using external GPX recordings.
As long as you use the Insta Remote to record the GPX track, the images and the position above ground remain synchronized. However, not everyone is happy with the number of images resulting in Mapillary. In Complete the Map Europe, I only achieve a very low value of 0.051 for images per meter. Furthermore, Mapillary images obtained this way tend to lose orientation when driving on the highway. The orientation then jumps around wildly. https://www.mapillary.com/app/user/osmplus_org?lat=48.25737887083298&lng=12.447665574999974&z=17&all_coverage=false&pKey=9740365596068184&x=0.7439054510348709&y=0.5734648762481929&zoom=0.005495261329194664&focus=photo I’ve now used an Osmand GPX recording for a short test drive. Look, I get more images per meter, and the orientation is stable. But there’s another problem. If I stop briefly because of pedestrians on the road, the GPX track and the image sequence subsequently diverge. https://www.mapillary.com/app/user/osmplus_org?lat=47.42417771250004&lng=11.750345225000046&z=18.89794144263493&all_coverage=false&pKey=1171387768357768&x=0.8390568357599029&y=0.5180863675881912&zoom=0.005495261329194664&focus=photo Does anyone have an idea how to solve this problem?
Look, there is no easy or automagical fix to all these problems, unless of course maybe somebody comes up with a clever AI solution but I would not count on it any time soon. Anyhow, for now you can also use JOSM or mapillary_tools to synchronize GPX tracks (‑‑interpolation_offset_time, ‑‑interpolation_use_gpx_start_time, and ‑‑offset_time) to your videos, cut sequences, and drop duplicates (‑‑duplicate_distance and ‑‑duplicate_angle), that is images captured when stopped. If you want to have high quality sequences then you are not going to get around putting in some manual effort into it.
@osmplus_org - something seems off with that calculation. Mapillary samples about 1 image every 3 meters (unless you’re going highway speeds, in which case it is more sparse). So ideally the number should be closer to 0.333 images per meter. If you look at your actual captures on Mapillary I assume you’re seeing the correct sampling rate of about 1 image every 3 meters?
1 Like
I am not sure what you are complaining about. To me this sequence’s positioning looks spot on. And yes, some positions may be slightly off due to speeding up and slowing down while driving but this is negligible and nothing out of the ordinary. Its density hovers around 0.333 ipm, which also is fine. 
Hence, I do not understand what you are complaining about.
@osmplus_org means that he achieved 0.051 ipm in the CompleteTheMap Europe contest, not in general. This number is feasible if his captures for the contest potentially had a high highway (high speed) percentage or low fps.
Just for the sake of clarity; ipm are irrelevant to the contest.
I had a lot of fun at the Complete Europe competition. I simply had a fantastic time. I’d definitely do it again. Perhaps it’s worth noting that the Insta isn’t the easiest camera to use. It’s a camera with a lot of potential, though. For beginners, I would definitely recommend the GoPro 360.
Here’s a comparison of two identical Insta360 x3 recordings: one using the Insta Remote GPX track,
and one using the osmand GPX recording.
The pleasing finding from this test drive: The osmand GPX track and the images did not diverge, despite traffic lights and a test stop during the journey. The number of images increased from 1,358 to 4,527 thanks to the osmand GPX track. This resulted in significantly smoother playback.
The unpleasant thing, which Mapillary might not care about, is that Google doesn’t like the osmand track because the time synchronization with the video is no longer 100% accurate. I’m not going to go to the trouble of correcting the time offset for Google, but everyone has to decide for themselves.
In any case, one more step is required. However, if I now forego ProRes422 resolution and limit myself to H264—which significantly reduces the number of video segmentations per half-hour ride from 5 to 1 video—everything speaks in favor of using the osmand GPX track recording from my smartphone as often as possible in the future. The Insta Remote is then just a remote control that also provides practical cutting marks.
2 Likes
Great! 
Looks like you found a better tool for your workflow.
1 Like
Not only that, by using OpenStreetMap data where there is no GPS reception, I can now also master tunnel passages without interruption. https://www.mapillary.com/app/?pKey=1073402984628753
1 Like
@osmplus_org - just to make sure I understand, are you saying that the .gpx track you recorded with osmAnd was able to record accurate GPS through a tunnel? Is this because you have OsmAnd connected to a ODB2 Bluetooth adapter?