Skip to content

Data folder

BioSegment works with datasets located in a ROOT_DATA_FOLDER folder, mostly a mounted network drive. Users can access the network drive directly to import or export files with their own tools and workflows.

Structure

  • ROOT_DATA_FOLDER
    • setup.json
      • JSON file containing configurations the backend reads during initialization of the database. Not required. Used in development. See the example at scripts/example_setup.json.
    • EM/
      • {dataset_name} e.g. EMBL
        • raw/
          • {pngs}
    • models/
      • {model_name} e.g. unet_2d
        • = output folder of neuralnets training
        • saved model e.g. best_checkpoint.pytorch
    • segmentations/
      • {dataset_name}
        • labels/
          • = ground truth labels of the dataset
          • {pngs}
        • {segmentation_name}
          • = output folder of neuralnets inference
          • {pngs}
    • annotations/
      • {dataset_name}
        • {annotation_name}
          • saved annotations e.g. annotations.json

New dataset

Adding a new dataset requires creating a folder with the dataset title in the EM/ folder e.g. {ROOT_DATA_FOLDER}/EM/{NEW_DATASET_NAME}. Then the dataset files need to be set in this folder e.g. a single .tiff or multiple .png files.