LargeScaleMeanShift - Is there a maximum file size?

Explain briefly what you are trying to do
I’m trying to get Large Scale Mean Shift to run on a 1.45GB dataset. It keeps crashing.

Configuration setup
My system**: Windows10
Version of the OTB
*: 7.2 Win 64
I installed the OTB with

Description of my issue
I’m trying to use Large Scale Mean Shift to segment an 8-band Worldview 2 image. The complete image is 1.45gb in size.

I have tried running the algorithm on a smaller area that I extracted and the algorithm works fine.

I want to apply the algorithm to the entire dataset but it keeps crashing.

Is there a maximum file size that this algorithm can be run on or shouldn’t this be an issue?


this algorithm should work on large images, this is why it is called “LargeScale” MeanShift.

Do you have any logs of the application before the crash ?

the LargeScaleMeanShift application is actually a combination of several smaller applications, as described here. Maybe you could try to run each of these step independently to find at which the the crash happens.



I get the following error message.

‘LargeScaleMeanShift’ has failed with return status -1. Please refer to ‘LargeScaleMeanShift’ documentation and check log tab.

The crash appears to happen in the ‘Computing Stats on input image’ stage.

The log states:

(FATAL) LargeScaleMeanShift: Caught std::exception during application execution: bad allocation

I’ve also just noticed these messages when initialising Monteverdi Application Launcher.

WARNG> Failed to load ‘qt_en_GB.qm’ translation file from ‘C:/OTB-7.2.0-Win64/OTB-7.2.0-Win64/translations’.
WARNG> Failed to load ‘en_GB.qm’ translation file from ‘C:/OTB-7.2.0-Win64/OTB-7.2.0-Win64/share/otb/i18n’.

Hi @wkcmark,
it seems be the same issue for me.
I tried with a small image and it’s working well. so i think there’s a size limite.

1 Like

@kadirim yes hopefully it can be fixed. I’ve been trying to get it to work since Monday.

My idea in the meantime is to cut the image down into smaller sections and process it that way. Not ideal but the only way I can think to carry out the analysis at the moment.

Yes this solution isn’t the best for me,
because when i merge the subsets segmentations i haven’t a perfect match on the edges and the limit it’s too much visible

@kadirim yes agreed.

@kadirim I think I’ve found a solution in the meantime.

There is a segmentation algorithm accessible through GRASS called i.segment.

You can access it using QGIS.

OK i will test.
Thank you

@kadirim Did you make any progress? Are you trying to carry out a pixel based classification or OBIA?