Some questions for those that do a lot of uploading with the tools

  • Is it safe to delete images after processing and before upload? This is quite common for me (today was removing those with rain wiper blades in the view) I note that each image (MAPPhotoUUID) has a defined sequence (MAPSequenceUUID) in the EXIF at this point, but my concern is the effect of removing from random places within the sequence.

  • I have a number of sequences uploaded that have lat/lon errors from an intermittent USB plug in the laptop. The error is actually a fixed amount in a particular direction, so I was considering a gpsbabel job to do a rough fix, then something like Viking to snap those remaining to the road. After saving that as a gpx I was going to rerun the tools, delete the offending sequence on the server, then upload again. Can anyone see anything wrong with this approach?

  • Is it generally safe to do sequence repairs by deleting them from the (Web GUI) server and rerunning the job?


I think so. Are you using the web browser upload tool for all your uploads?

1 Like

No I am using mapillary_tools from the Linux command line. There is also a delay of between one week and two months between process and upload.

1 Like

Hi, have you managed to do the job?

I am having similar issue - during several hours continuous recording my dashcam is loosing gps fix for 3-10 minutes sometime (I suspect buffer issues or gps chip overheating).
Also this is common issue while driving tunnels (i.e. driving in Alps)

This results in part of the sequence not geotagged properly.
After running a process job, all images on this part of the sequence are being geotagged on a straight line (between last known location and a first location where fix was restored)

I can use Geosetter to extract resulting geotag data to a gpx with trackpoints

Resulting gpx file will have a correct datetimestamp (matching image samples) but erroneous GPS lon/lat

I wonder if there is a way to snap this track back to a road, preserving at the same time:

  1. number of track points (should match # of samples for re-processing)
  2. date/time data of each point - they should stay as is, i.e. 1 trackpoint per second
  3. distance between the points should be distributed evenly (assuming constant driving speed)
  4. GPS coord should be updated accordingly (snapped to road I choose)

I have checked Viking GPS tool but it does not seem capable of doing this…
All I have found - is different tools (i.e. Routeconverter) where I could build a route along the road (between two or more points) and convert it into a trackpoint list (additional positions are added automatically to keep it aligned with the route).
But there are no settings/controls to set sampling rate (# of trackpoints), start/end date/time, force even distribution (equal distance between each trackpoint)

It is safe to delete some of the ‘processed’ images prior to uploading.
It will just make the resulting sequence a bit smaller.

However I have a different question related to sequences.
Once process command is finished, it stores geodata and mapillary metadata in image EXIF headers.
This includes sequence ids assigned during first processing.

Sometime I am running processing on a subsets smaller than 500 images.
Let’s say, I have processed 3 sets, 100 images each.
Running the processing separately assigns each of the subsets a unique sequence id (=3 sequences, 100 images each).

Now I want to join 3 subsets into 1 bigger sequence (i.e. 300 images)
If I will re-run the process - it will just take the sequence ids from the meta without actually updating them (without joining).
I was struggling to find a mapillary_tools option/command to force re-sequencing but failed.
Does anyone have ideas, is it possible with mapillary_tools?

The only workaround I have identified was - to use Geosetter to clean the mapillary metadata from Exif, and than use web uploader to upload the resulting sequence. Maybe tools process will do the job as well after metadata was removed from the images…

No I didnt try to correct the GPS error. I will be revisiting the site soon and capture with a better camera. (That same GPS unit now has the 1024 week rollover issue to…)

I did think through just doing some maths on a gpx file. It would be quite easy to read a lat/lon and modify it by a distance and bearing. Only useful for a fixed error though. I suspect that it would be a relatively easy to do some route vs track maths in a spreadsheet. eg pick the nearest route point from each of the (error) track points and output that. I assume it’s reasonably easy to create a route file for all roads (from OSM) in a particular area, to use as the source.

1 Like

I managed to overwrite the Mapillary EXIF data when I had auth issues, by deleting the .mapillary/log subdirectory and doing the rerun as if it was a image set rather than a movie. (I had written EXIF headers as part of movie processing) I wasnt actually looking to change sequence info though.

I notice too that there is a sequence/split algorithm that looks at time and distance between points, that could be modified.

indeed, this could work, I will give it a try out of curiosity :slight_smile:

This was the first thing I’ve tried. But I’ve noticed that original sequence ids were still preserved (taken from existing EXIF meta) and not re-calculated from scratch.
So I was forced to clean the Exif mapillary meta first with Geosetter and process again so the new sequence ids are assigned.