Request for Changes-101: Optimization of SampleExtraction

From OTBWiki
Jump to: navigation, search

[Request for Changes - 101] Optimization of SampleExtraction


  • Author: Guillaume Pasero
  • Additional Contributors (if different than authors)
  • Submitted on
  • Proposed target release
  • Adopted / Rejected (with vote results)
  • Branch : optim_sample_extract


This RFC introduces optimization in the sample extraction process.


The sample extraction step is often performed after having computed a set of features on the initial image. This feature extraction step can be demanding in computation time. If there are areas where no sample has been selected, then these areas can also be skipped when computing classification features.

Implementation details

Classes and files

M       Modules/Learning/Sampling/include/otbImageSampleExtractorFilter.txx
M       Modules/Learning/Sampling/include/otbPersistentSamplingFilterBase.txx

The main idea of this optimization is to skip tile where there is no sample, and also to compute the useful extent of non-empty tiles

The changes made to the base class otb::PersistentSamplingFilterBase should allow to detect empty tiles (tiles without selected samples) during the GenerateInputRequestedRegion() step. This way, when an empty tile is found, the requested region for the feature image is set empty.




Additional notes