code/scripts/h800_cache_data.sh

#!/usr/bin/env bash
# Cache the dataset from the slow JuiceFS share to local RAID /data/temp for fast
# training reads. Parallel per-dataset cp -a (preserves pannuke hard links), overlaps
# JuiceFS small-file read latency. Run detached; log to /tmp/cache_data.log.
SRC=/mnt/tidal-alsh-share2/dataset/qinshengqian/research/c3/NPJ-ACM/Data
DST=/data/temp/NPJ-ACM/Data
mkdir -p "$DST"
echo "[start] $(date +%T)"
for d in acdc_png busi cvc_clinicdb fives idridd_segmentation kvasir_seg \
         medsegdb_isic2018 medsegdb_kits19 pannuke_semantic refuge2; do
  ( cp -a "$SRC/$d" "$DST/" && echo "done $d $(date +%T)" ) &
done
wait
echo "CACHE_DONE $(date +%T)"