Big data computing (photogrametry) - Processing Time

Answered

Comments

12 comments

  • Avatar
    Jonathan_Tanant

    Hello,

    I think that the first thing to do is plan the processing. You will not be able to put 640,000 pictures in the software and expect something to come out.

    How are pictures organized ? What is this exactly ?

    The main idea on big projects is to align small overlapping clusters seperately and then align clusters together, and clusters of clusters until you have the alignment output you need... And this require precise planning. You can do this manually and/or with the CLI.

    Still 640,000 is quite a high number of pictures !

     

    0
    Comment actions Permalink
  • Avatar
    Jefferson Moreno

    We appreciate your quick response.

    We obtained the following information from RC, which tells us that we require 128GB of RAM to process 16,000 photographs but they do not specify how much time is required (because they do not know the hardware performance).

    Jonathan, what we have thought so far is to separate the 640,000 photographs in 20 different lots, which means that we would need 20 machines working at the same time. * 640,000 / 32,000 = 20

    The above means that we need at least 256GB RAM per machine to proceed with processing. But we do not have information about the necessary time, and we would appreciate if you have more information about the planning that you mention.

    0
    Comment actions Permalink
  • Avatar
    Jonathan_Tanant

    I don't understand. What are your pictures subject exactly ? Is that aerial shots ? is that separate objects ? is that only one object ? what do you want to do at the end ? Please provide some details and an example if you want some feedback.

    You can't just split your pictures in 20 random or sequencial lots and "process" them. It does not work like this. 

    0
    Comment actions Permalink
  • Avatar
    Jonathan_Tanant

    And from my experience, trying to align 32,000 pictures in one run will not work and will not be efficient at all - even on a big machine.

    Of course it depends on what your datas are.

    0
    Comment actions Permalink
  • Avatar
    Jefferson Moreno

    All right, we'll be more specific.

    The images are in.jpg format and the quantity is due to the fact that we have to cover an area of approximately 100,000 hectares using drones in order to perform photogrammetry.

    Actually the number of photos is 300,000 to 400,000 and they correspond to one of the largest projects we have pending.

    The idea then is to work with smaller projects of around 100,000 photographs. Thus, our goal at the end of each project is to obtain a 3D model of the terrain covered by the drones in a record time.

    Jonathan, finally, could you tell us from your experience how long it has taken you to work with 32,000 photographs?

     

    Once again, thank you for your cooperation.

     

    0
    Comment actions Permalink
  • Avatar
    MikeS11

    I would for sure just cut the work into pieces and run them across multiple processes just as suggested.

    Are you Google ? Haha

    0
    Comment actions Permalink
  • Avatar
    Jonathan_Tanant

    Time depends on the project and what you want to get as an output, not only on the pictures count.

    What I would do :

    -generate 20% to 25% overlapping clusters of ~1,000 pictures. You can do squares of your terrain. You will have around 800 clusters to process for each group of 640,000 pictures (approx.).

    -process them individually (first manually to test, then with the CLI) and export registration for each of them. It may be hard to know exactly the time needed to process 1 cluster (it depends on many factors related to the number of features the software can extract from your pictures) but I would say less than half an hour each on a high end machine.  So maybe you will need 400 hours (16+ days) to align all your pictures. It could be more or less, it really depends on your data. 5 MP is not much, so maybe I overevaluate the time - maybe 10 minutes or so will be enough. Just run a test.

    -Then you import all your registrations (the 800) in one project and you align again to (try to) get one only alignment. It should take some time, maybe days - once again, hard to tell without some testing.

    -Then you reconstruct your model. Depending on the level of precision, you have a choice to make between fast reconstruction and accuracy (with the downscale for depth map setting).

    -Then simplify, cleaning and texture.

     

    Hope this helps.

     

     

    1
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    2-3 weeks? Phew.

    And 3 TB SSD? That is likely a bottleneck because I think it doesn't even quite hold all the images, right?

    I don't want to imagine the size of the sparse point cloud let alone the depth maps! Just shooting from the hip I would say that you probably need a much bigger SSD (HDD) and also several (for temp folder, source images and project files).

    Let us know if and when you managed that feat!  :-)

    Good luck!

    0
    Comment actions Permalink
  • Avatar
    Götz Echtenacher

    Plus I recommend getting in touch with support directly...  :-)

    0
    Comment actions Permalink
  • Avatar
    Lucia CR

    2 million terrestrial photos, 2.5 milion aerial images: https://www.youtube.com/watch?v=QSqZyvKhw3U

    by: https://www.youtube.com/channel/UC6fjDI-mHkAZT9Pnr29hg-w/videos

    0
    Comment actions Permalink
  • Avatar
    Jefferson Moreno

    Hello again, thank you all for your time.

    @Mike Simone- haha we are not Google yet, but we are trying to solve a deforestation problem in Colombia.

    @Jonathan_Tanant - Thank you a lot, that is a very interesting starting point. We are going to start our tests using RC and Google Compute demos.

    @Götz Echtenacher - We forgot to say that we are thinking in an 20TB of cloud storage for the information, but as we have understood, each machine will be processing different sized clusters, and not the 640k photographs at once. *We sent an email to RC and they told us that it was best to ask in the forum.

    @Lucia CR - Thank you for that example, It means that it is possible to carry out large projects using the RC software. however, our challenge for now is to do things in a short time.

    Thank you all guys for the information, we will be in contact to keep you informed of our progress.

    0
    Comment actions Permalink
  • Avatar
    jerome.c.leray

    @Jefferson Moreno, How did you go?

    0
    Comment actions Permalink

Please sign in to leave a comment.