Status on large change sets and change set process


#1

For some time it has not been possible to process change sets (e.g. hide images) larger than 100 images. Is there a status to when this number goes up? I have made a few mistakes I would like to correct :smile:

Also, for small change sets (1-20 images) I have made submissions to hide images, but they are still there 2-5 weeks after the submission. Are the requests processed automatically or does a human have to approve? I have never received a confirmation or rejection mail on change requests, so I have never really trusted the feature.

I some times get approvals of blur requests several weeks after I made them, so I guess there is a human in the chain here?


#2

I can tell there is human validation for every request. If you have enabled the option in your profile settings you should receive and email with a title like “We have successfully registered your blur requests” a few minutes after submission. Then an email titled “The blur edits you requested were approved” will come a few hours/days later (depending on the workload and availability of the validation team).


#3

Thank you @LeGrandBlond, that leads to 2 questions:

  1. If there is human validation on a change set like hiding an image or sequence, how can that human know what to accept and what to reject? I have tried to hide a sequence because it was geotagged to get wrong coordinates. Everything was perfect except for the location, but only people who had been on the ground could know that. I have also tried to hide images with wrong date (2013 instead of 1960). There is no option to comment why you want something hidden, so it is difficult to tell.

  2. I have the email option enabled, and I get the emails for blur requests just fine. But when I hide a sequence or an image I have never received an e-mail.

I have to problems:

  1. I have made a lot of hide requests and I think a lot of them (if none at all) have been accepted. I have not made a note of them, but those I can remember for sure have not been accepted, even months after.

  2. I have made geotagged some sequences wrong (daylight savings time), so I need to hide some big sequences. But with a limit to 100 it would take days to hide the images individually.


#4

I guess human validation is to prevent vandalism, mistakes, and privacy concerns. Someone deleting his own photos should definitely go through.

I agree with you @tryl , the “limit to 100” is a real pain to fix bugged sequences. @Sandra , is this problem being addressed ?


#5

That number 100 is not a hard number; they just mean to say the process is having problems dealing with large change sets. The current warning that is shown is from my suggestion, the previous warning was even more unclear. See also issue https://github.com/mapillary/mapillary_issues/issues/1315. However, it seems that currently no changes are processed at all anymore (though I can not find the issue number anymore, where this was mentioned).


#6

Hi @pbb and all,
The issue where this was mentioned is https://github.com/mapillary/mapillary_issues/issues/1367


How to delete a faulty sequence?
#7

Thanks. I remember having seen that issue, but it is a long time ago there have been any activity.
Perhaps @peter or @gyllen could give a short status?


#8

Hi there,
sorry for being quiet on this topic for a while. Status is that the changeset structure is not optimal right now, so we are doing rework there. We ARE approving blurs, that is why they are processed. However there are 2947 unapproved changesets that will need to be treated. Pretty hefty backlog, but we are thinking of auto-approving changes to your own images.

The current status is that we have been working and coding on support for detections and object merging in the backend - having largely a similar, incremental trackable structure as changesets. In 2 weeks @gyllen is gonna come over to Sweden with basically the single focus on working on this. After that we can at least start approving and hopefully remove the soft limit of 100 changes (that is due to the RDBMS being slow in processing large changeset inserts).

So, we hope to have a XMas present for you!

/peter


#9

Can’t the status message that is currently shown on the edit page, be updated to inform that currently no changesets (apart from blur requests) are being processed?


#10

I have a problem with one of my sequences. I added a video to the wrong GPX file. Is there a way to mark a complete sequence to be removed so as not to edit each photo and hide (this still not work correct)?


#11

Under sequence settings in the edit menu, there is a button ‘Hide whole sequence’, but as @Peter already stated are there many changerequest waiting to be aproved, so it might not work directly


#12

Hi!
Any news on that topic @peter?
Regards, Markus


#13

Yes.
I’m actively working on the new changeset structure. Right now, the structure for location changes is in place, we are going to deploy the code after the holidays and and start converting the old changesets to this structure, and build the audit infrastructure to approve them.

Sequence changes, project changes and image deletions/hide will follow when location changes are working. Stay tuned, sorry for it taking longer but we have also some data cleanup to be done in the process so more of the work can be done by Elasticsearch instead of Postgres which can’t handle the amount of changes and traffic we are having today.

/peter


#14

Good to know, thanks @peter.

These are good news, somehow, to learn that the good old PG does not handle the load anymore. Price of success :smile:

Sure you’ll have better perfs with Elastic. It’s based on Lucene/SolR if I’m not mistaken?


#15

Yup,
Elastic search is scaling out Lucene indexes under the hood. Coming back to work on Monday, changesets on the way!

Happy New Year!


#16

Do you have any update?


#17

Hi there,
yes. we are at the same time working on merging object detections into objects, and the changesets. As of yesterday, the basic server side structure is working - we are just writing more tests to make sure the new changes to the codebase are not overwriting valuable information. Starting tomorrow, I can then test the conversion of old changesets to new ones and we will start on the UI to approve them. Things are moving, sorry it takes a long time!


#18

Take the time necessary to do to right. We will try our best to break what you are making, by uploading huge amounts of images, so better spend the extra time than sticking to some old deadline.


#19

Thanks for your understanding - we are basically in the process of writing more tests for the Apache Storm Topologies we are moving a lot of code into (they are insanely scalable, but you have to make sure you end up with the right data written to the databases and have the right cached data to make decisions on, http://storm.apache.org/documentation/Trident-API-Overview.html). Not the most easy stuff I ever read :smile:

/peter


#20

Any news on the change set processing, @peter? It does not seem like anything in the edit interface is functional at the moment.

I think it would be best to remove access to the edit interface completely, for the moment, since it’s non-functional.


The sequence editor makes me nuts!