Skip to content

Data augmentation bug

Hélène Touchais requested to merge data_augmentation_bug into develop

When you want to use data augmentation with a data_field containing a "_", otb returns the following error.

RuntimeError: Exception thrown in otbApplication Application_SetParameterString: /iota2/miniconda3/conda-bld/otb_1655497468350/work/OTB/Modules/Wrappers/ApplicationEngine/src/otbWrapperListViewParameter.cxx:127:
itk::ERROR: FieldParameter(0x7f2326e36e50): Cannot find data_field

The solution implemented here is to use the i2label field for data augmentation instead of the data_field from the configuration file parameter.

Another error can occur here when labels (data_filed) are of type str. A type check has therefore been added.

In addition, we add a check to prevent the use of data augmentation when gap-filling is disabled.

We also addressed the issue #639 :

To add the possibility of using data augmentation with Pytorch, the following modifications have been made:

  • Generation of the training samples SQLite file in the case where data augmentation is activated.
  • At the end of data augmentation step, generate a netCDF training file from the augmented SQLite file and delete the SQLite file.
Edited by Hélène Touchais

Merge request reports