skysat_overlap.py stuck at saving bound shapefile

[ShiruiH]

Hi Shashank,

Thanks for composing this great tool for processing SkySat data.

I've encountered an issue when running skysat_overlap.py. The number of images I'm processing is 966. I understand this will generate 466095 combinations. When the script is saving the bound shapefile to my predefined directory. It is extremely slow in processing and always gets stuck during the process (usually stopped at 3% - 5%). I've tried changing the directory, but the issue remained unsolved.

I'm just wondering if there is any solution in your mind that can resolve this issue?

Cheers,
Shirui

[ShashankBice]

Hi Shirui,
Thanks for reporting this. I can see why this procedure is a bit slow, the logic right now is brute force and just spread across cores, which will make it slow if number of cores on a machine is low. How many cores does the machine you are using to run these scripts have?
Regardless, I get much faster performance when finding intersection using the native rtree algorithm in geopandas for another related project, which I will port here in a day. I will let you know.

Cheers,
Shashank

[ShiruiH]

Hi Shashank,

Thanks for your prompt reply. I have 14 cores available to be used on my machine.

Looking foreword to your new solution with the native rtree algorithm.

Cheers,
Shirui

[ShiruiH]

Hi Shashank,

May I know if you're running this process on a HPC with more cores? Or if you may have any update with using the rtree algorithm to find intersections?

I can see that from your code, every two images are paired up with a for loop going through the whole image list to find their intersection with the gpd.overlay() method. Are you proposing to substitute gpd.overlay() with rtree? Will you keep the for loop?

Cheers,
Shirui

[ShashankBice]

I am sorry for my long silence, I updated the function using rtree for finding intersection, and it produces same results as the earlier function with ~13x improvement in execution wall time. If you installed the package as editable, these should be visible after a git pull.

Let me know how it goes,
Cheers,
Shashank

[ShiruiH]

Hi Shashank,

Sorry for the late reply. Thanks for improving the package, it is much improved now!

I'm now at step ba_skysat.py. Running into an error at line 308:
IndexError: index 1 is out of bounds for axis 0 with size 1

I think the reason behind it is that in my case, img_time_unique_list only has one element (sc00103), so it failed when it tries to access img_time_unique_list[1].

So I just wanna confirm with you if string like "sc00103" is the img_time you intended to record? My files have names such as "1231553292.98343730_sc00103_c3_PAN_i0000000006". Apologise if the question is trivial!

Cheers,
Shirui

[ShashankBice]

Hi Shirui,
Looking at the filename, I suspect this is a L1A full frame triplet stereo product? Please let me know. At this stage, we do not have scripts for L1A full frame product processing, although we have the logic, and we will come up with it in the coming month or so. In the meantime, I added a note in the project readme to mention the currently supported products .