Skip to content

README.md

Latest commit

 

History

History
42 lines (31 loc) · 1.36 KB

README.md

File metadata and controls

42 lines (31 loc) · 1.36 KB

Data Preprocessing

This directory contains preprocessing scripts for all datasets used in DeepSparse.

Structure

data/
  base/           Shared utilities (projector, dataset, saver, utils)
  atlas-mini/     Pretraining dataset (AbdomenAtlas1.0Mini)
  PANORAMA/       Finetuning — abdomen (PANORAMA challenge)
  PENGWIN/        Finetuning — pelvis  (PENGWIN challenge)
  LUNA16_v2/      Finetuning — lung    (LUNA16)
  ToothFairy/     Finetuning — tooth   (ToothFairy challenge)

Shared Utilities (base/)

  • projector.py — TIGRE-based cone-beam CT forward projector
  • dataset.py — Generic Dataset class: resample → crop/pad → normalize → block conversion
  • saver.py — Saves processed CT, blocks, and projections with consistent folder structure
  • utils.py — SimpleITK load/save helpers

Dataset Naming in Training

The dataset_img_dir_dict in code/datasets/base.py maps training dataset names to these folders:

Training name Folder
atlas-mini atlas-mini/
abdomen PANORAMA/
pelvis PENGWIN/
luna LUNA16_v2/
tooth ToothFairy/

Dependencies

pip install SimpleITK scipy numpy tigre matplotlib tqdm

See each dataset's README.md for dataset-specific download and preprocessing instructions.