#!/usr/bin/env bash # # Install medvision_ds so build_dataset.py can run. # # ./setup-env.sh /path/to/data-dir # MedVision_DATA_DIR=/path/to/data-dir ./setup-env.sh # # medvision_ds is installed through medvision_bm, which is the documented public route. # Nothing here is specific to any machine: no absolute paths, no interpreter pinning, no # credentials. Run it from any checkout. # # WHICH medvision_ds YOU INSTALL DECIDES WHAT GETS STAMPED. build_dataset.py names the # annotation version explicitly when reproducing, so the installed version does not # affect the filename in that mode -- but it does decide the CODE that computes the # values. To reproduce a published annotation faithfully you want the released package, # which is what this script gives you. To build with local modifications, install the # working tree instead: # # pip install -e "/src" # set -euo pipefail DATA_DIR="${1:-${MedVision_DATA_DIR:-}}" if [ -z "${DATA_DIR}" ]; then echo "usage: $(basename "$0") (or export MedVision_DATA_DIR)" >&2 exit 1 fi if python -c "import medvision_ds" 2>/dev/null; then echo "medvision_ds already installed: $(python -c 'import medvision_ds; print(medvision_ds.__version__)')" echo "Nothing to do. Uninstall it first if you want a different version." exit 0 fi pip install medvision_bm python -m medvision_bm.benchmark.install_medvision_ds --data_dir "${DATA_DIR}" echo echo "medvision_ds $(python -c 'import medvision_ds; print(medvision_ds.__version__)') installed." echo "Next: python build_dataset.py --data_dir \"${DATA_DIR}\" --dataset --dry_run"