segmentation fault in otbcli_TrainImagesClassifier with a high number of input images
In my tests, otbcli_TrainImagesClassifier works fine with three Sentinel2 input images (tiles 10000x10000 pixels). However with 11 input tiles, I get a segmentation fault at the end of the SampleSelection step, and just before the SampleExtraction step :
[node039:09908] Signal: Segmentation fault (11) [node039:09908] Signal code: Address not mapped (1) [node039:09908] Failing at address: 0x10 [node039:09908] [ 0] /lib64/libpthread.so.0(+0xf100)[0x2b11b42ef100] [node039:09908] [ 1] /work/logiciels/otb/rh7/develop/bin/../lib/otb/applications/otbapp_SampleExtraction.so(_ZN3otb7Wrapper16SampleExtraction18DoUpdateParametersEv+0x1f4)[0x2b11ca8c33b4] [node039:09908] [ 2] /work/logiciels/otb/rh7/develop-20171218-080535/lib/libOTBApplicationEngine-6.3.so.1(_ZN3otb7Wrapper11Application16UpdateParametersEv+0x5d)[0x2b11ab3f0e6d]
I tried with different values for the ram parameter (10000 and 100000) and the segfault always occurs.
Tests were performed on HAL (CNES cluster), using otb/6.2 and otb/develop (same problem for both versions). Classifier parameters are: classifier rf -classifier.rf.nbtrees 100 -classifier.rf.max 20 -classifier.rf.cat 16