System is a vandal

There have again been untold improvements in the Mapillary Explorer.
Now there is a little picture of the edits.

Yesterday I took a picture of me pushing the button of the drinking water fountain to prove that it still works. System deleted it because it is no streetlevel imagery.

I take this very seriously.
I hope this is no policy and just a bad judgment of a “trainee”.

Am I the only victim ? Are there other reasons why people left ?

I dont think I have ever had anything deleted (other than ones I marked for deletion). Load it again and see what happens.

Still lots of sequences in Parks. Was it just a single image (ie one image in the sequence)?

What did support@mapillary.com have to say about it? I know they dont actively monitor these forums so I usually send queries there if its urgent or I want specific answers. I would also send a URL of where the image was deleted.

It is only the second time. The previous time I don’t know what they edited.
There was one picture of my arm pushing the button, but they deleted the whole sequence in the park.

I am very interested if I run the risk again. I like to make a picture of a mushroom from time to time when I cover a path in the woods.

Is it some Facebook sweatshop that judges ?

I prefer to make things public before I contact support.

A part of the sequence in the park is slowly appearing, but not on the map yet.

Nevertheless, in 3 months I will not know anymore if the fountain worked.
This objection is just a matter of principle, because all fountains are in corona lockdown now.

This is just an example of the bad manners of Mapillary that we have become used to.

It appears indoor AED’s are no longer accepted either as ‘not street level imagery’. Shame.

This is the end of Mapillary.

We have a review system in place that reviews uploaded imagery that our algorithms flag as not street-level imagery. A human then reviews these flags and either:

  1. Approves them - imagery is deleted
  2. Rejects them - Imagery remains

This is actually a big improvement because we’re trying to make sure that images on the platform are useful for mapping purposes.

@filipc the image you describe would not meet that criteria. If the image was just of a water fountain then that’s different.

@TimCouwelier the vast majority of indoor imagery we see is not suitable for Mapillary. Usually it’s people testing the app or unaware that they were capturing imagery. An AED indoors is in fact useful though. Please flag future examples of this so we can be sure to retain these images.

1 Like

Glad to see. I’m however not able to retreive a link for the edits - they were requested on 2020.07.06, and edited by @system. Was a set of AED images all over the country - often have to take these by normal camera app as Mapillary app is picky on GPS location indoors. Once every few months, when cleaning pictures on my phone, I copy the AED ones and upload them with the desktop uploader.

  1. User makes sequence
  2. Secret human thinks picture NOK
  3. Sequence is deleted

@eneerhut Could you please clearly define what is meant by “mapping purposes”? If Mapillary now starts to reject images, it would be good to know exactly what the criteria are.

Indoor imagery can be very valuable to OpenStreetMap, for example from inside a shopping mall, showing the different shops, or inside a public library, showing the location of stairs and lifts.

Obviously, if Mapillary is more interested in the value of images for vehicle guidance systems, then I understand that these are not interesting to Mapillary, but it would be reasonable to be clear about that.

3 Likes

You’re right, it is worthwhile getting some clear guidelines on the help pages.
We’ll coordinate to get this together soon.

The initial purpose of this tool was to prevent unhelpful imagery. While “unhelpful imagery” can be ambiguous, there has long been an implicit understanding that Mapillary is a platform for street-level imagery used to derive map data. Therefore imagery of people or indoor environments is discouraged.

Mapillary is not well set up to handle indoor imagery. Firstly, because GPS does not work well indoors, so images need to be manually placed. Secondly, because Mapillary does not distinguish between indoor building levels, so there is no mechanism to give the vertical location of an image in a building with multiple levels.

That being said, I can see the real value in use cases like AED mapping that was mentioned.

A good starting point is for us to clearly define what images are unacceptable on Mapillary, in addition to what has already been described in our terms. We can then continue to discuss some of the ambiguous situations at the margins.

2 Likes

As long as my mapping example is not unambiguosly acceptable, it makes no sense to work for Mapillary.
A system should be in place to warn the contributor and to ask his opinion.
In case of doubt -> the customer is always right.
By the the time Mapillary sees the light and a system is in place, a summer will be lost.

https://www.mapillary.com/manifesto

2 Likes

@filipc this is an error on the side of the reviwer of images that are flagged as potentially being non-streetlevel images. I will try to find and un-delete them. We are taking vandalism even more seriously now and create changeset generated by algorithms that suggest images that look like non-suitable. These changeset are then approved/rejected by a human (in this case falsely approved) every day.

Over time this will of course get better adjusted, for now this is mostly due to the reviewer interpreting the changeset wrongly. I will revert, sorry.

/peter

1 Like

I’ve uploaded many images, often requiring a lot of extra effort, that are useful in mapping.
Opening hours. Some indoors images. Paths in the woods.
If images like @filipc’s water fountain one are suddenly considered not desirable, I would lose all motivation to contribute.

1 Like

Errors happen.
Deleting without asking is not an error, it is an attitude.

As mentioned - this is not an attitude change but a human error approving a flagged deletion-changeset which was wrong. No auto-approving anywhere.