### Description Sample extraction application memory usage is way above the value set by the RAM parameter. Looking at full dev messages, I can see that pipeline memory estimation is coherent with loaded image data, but the application memory usage is from twice to 10 times the estimated value. Worse, the application memory footprint grows constantly as it processes chuncks of input images. I think all this can not be explained by the extra memory from the output vector features created. ### Steps to reproduce Any use case of sample extraction, but larger images with larger RAM parameter make the problem more obvious. I can provide sample data if needed. ### Configuration information Linux (RedHat 7) with up-to-date develop branch, but should not be relevant for this bug.