• pip install --upgrade git+https://github.com/MakiResearchTeam/MultiClassLabelBalancing.git

General steps

1. Perform dataset analysis using HCScanner:
   1. Generate .npy file containing information about labelsets (labelsets, alphas).
2. Perform balancing using available algorithms:
   1. Generate new .npy file containing information about labelsets (labelsets, alphas), but with new alphas.
3. Do construction of a new dataset:
   1. You can do copying of existing data using new alphas and then perform uniform sampling from the resulting dataset.
   2. Or you can build a generator that would sample pictures containing certain labelsets using a distribution in order to sample them.