sarproz-group

Problems in the APS processing

This topic contains 12 replies, has 2 voices, and was last updated by  stevexu 2 years, 2 months ago.

  • Author
    Posts
  • #1061

    stevexu
    Participant

    It seems the APS step has trouble. I will put all the screen shot of the setting, result, statistics. Need help for re-run to get modification.

    1. The cumulative height of the PS seems has a systemic drift towards big value, which gather above 800m and up to 1000m

    2. I check the simulate DEM, it seems all right

    3. I tried also the linear-trend instead the smart one, result is even worse.

    Attachments:
    You must be logged in to view attached files.
  • #1069

    periz
    Keymaster

    you need a good amount of patience and experience to become able to play correctly with the parameters and to be able to judge what is wrong in the results.
    I do not know the details of your dataset and of the area you are processing, so, I cannot give precise suggestions.
    Also, which images graph did you use?
    Anyway,
    a. why are you setting DL=30. I suggest you to leave DL=0
    b. whatever is your dataset, with 15 images you cannot leave the height range -100 100. If it’s a non-urban area, if you are using SRTM you should be able to restrict it to something like -30 30
    c. your consideration on using the linear deformation trend was biased exactly by the too big ranges. If you can guess the expected deformation, highly probably with 15 images the linear def works better than the smart option (not sure anyway, it depends on the signals u have). In such case, you have to strongly reduce the velocity range. Also considering that u are using the delaunay graph, probably -5 5 or -10 10 is enough.
    There are many factors to consider. A way for you to take confidence is using the small area and test many different configurations. When you get familiar, you can use the best configuration in the APS processing…

  • #1072

    stevexu
    Participant

    Yeah, you are quite right, I reset the parameters as your recommendation, in which:

    1. DL=0;
    2. Linear trend

    I am using a dataset mainly located in the urban area, I re-run the “Sparse Point Selection” and “Graph Creation”, but stopped at “Connection Processing”, the runtime window shows as below

    In fact, I do this step in last evening, and till this morning the status is still 0%, I re-run the Sparse Point and Graph Creation step

    Attachments:
    You must be logged in to view attached files.
  • #1074

    periz
    Keymaster

    the parallel processing waitbar requires a network protocol to exchange infos between workers
    probably this communication was lost in your system and the waitbar got stuck, blocking the processing.
    you can check it if the waitbar does not move and there is no matlab instance keeping the CPU busy.
    In such a case it should be enough to restart matlab (or the application).
    It’s a good habit to keep an eye on the CPU usage to check that everything works fine.

  • #1075

    stevexu
    Participant

    Thanks for you attention, the Matlab parallel pool is still busy, but waitbar is still 0%.

    Attachments:
    You must be logged in to view attached files.
  • #1077

    stevexu
    Participant

    Here is the Matlab windows

    Attachments:
    You must be logged in to view attached files.
  • #1079

    periz
    Keymaster

    how many PSC/Connections are you processing? Also, what is the size of your dataset (samples and lines)? is your data on a local or external disk?

  • #1081

    periz
    Keymaster

    if Matlab says it’s busy, that’s not enough to understand if it is calculating things or if it’s waiting for something.
    You have to use the task manager in windows to see if there are 4 Matlab parallel processes at full work. Otherwise top in Linux.

  • #1082

    stevexu
    Participant

    I use top, only one matlab process is running, as describe below:

    That was strange, it has occupied only 5% of CPU

    Attachments:
    You must be logged in to view attached files.
  • #1101

    stevexu
    Participant

    I crop the area smaller, which up to 5000 * 6000 pixel, but it still frozen there, since I set the DL=0, the top command shows only one eternal occupied process is by MATLAB, though somehow parallel MATLAB workers show sometime.

    The graph processing has been running for up to 50000 seconds, I am playing on 4 core 16 GB RAM machine, is that suppose to happen?

    Attachments:
    You must be logged in to view attached files.
  • #1103

    stevexu
    Participant

    I forgot to mention, the PS numbers and connection is in the screenshot below

    Attachments:
    You must be logged in to view attached files.
  • #1105

    periz
    Keymaster

    As I said, it’s a network problem.
    Did you resart Matlab?
    Check hostname and IP address. There is a conflict somewhere. The workers try to send signals to the main Matlab instance, but it’s addressed in a wrong way and so the parallel job freezes.

    • This reply was modified 2 years, 2 months ago by  periz.
  • #1108

    stevexu
    Participant

    You are quite right, I did not discern the host address is changed due to the wireless internet reconstruction in our department, now I changed to the wire internet, it works smoothly. Thoroughly thanks!

You must be logged in to reply to this topic.