dvc.lock

schema: '2.0'
stages:
  data_collection:
    cmd: python src/download_dataset.py
    params:
      src/config.yaml:
        dataset:
          folder: src/data
          species_folders:
            Abies_alba: data/imagery-Abies_alba.zip
            Abies_nordmanniana: data/imagery-Abies_nordmanniana.zip
            Castanea_sativa: data/imagery-Castanea_sativa.zip
            Fagus_sylvatica: data/imagery-Fagus_sylvatica.zip
            Larix_decidua: data/imagery-Larix_decidua.zip
            Picea_abies: data/imagery-Picea_abies.zip
            Pinus_halepensis: data/imagery-Pinus_halepensis.zip
            Pinus_nigra: data/imagery-Pinus_nigra.zip
            Pinus_nigra_laricio: data/imagery-Pinus_nigra_laricio.zip
            Pinus_pinaster: data/imagery-Pinus_pinaster.zip
            Pinus_sylvestris: data/imagery-Pinus_sylvestris.zip
            Pseudotsuga_menziesii: data/imagery-Pseudotsuga_menziesii.zip
            Quercus_ilex: data/imagery-Quercus_ilex.zip
            Quercus_petraea: data/imagery-Quercus_petraea.zip
            Quercus_pubescens: data/imagery-Quercus_pubescens.zip
            Quercus_robur: data/imagery-Quercus_robur.zip
            Quercus_rubra: data/imagery-Quercus_rubra.zip
            Robinia_pseudoacacia: data/imagery-Robinia_pseudoacacia.zip
          main_subfolders:
            aerial_imagery: imagery/
            lidar: lidar/
        device: gpu
        model:
          name: fine_grained
        training:
          batch_size: 32
          learning_rate: 0.0001
          max_epochs: 100
          freeze: true
          weight_decay: 0.0001
          step_size: 4
          gamma: 0.1
          oversample:
            oversample_factor: 4
            oversample_threshold: 1000
          dataloader:
            auto: true
            num_workers: 0
            pin_memory: false
            presistent_workers: false
          early_stopping:
            apply: true
            monitor: val_loss
            patience: 3
            mode: min