run_lightly_distill_dinov3_splus_vitb.sh

#!/usr/bin/env bash
set -euo pipefail

ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

CUDA_DEVICES="${CUDA_DEVICES:-0,1,2,3}"
CONDA_ENV="${CONDA_ENV:-deimv2}"
DATA_DIR="${DATA_DIR:-/home/nikhileswara/DEIMv2/data/crowd_human_mot_coco/images/train}"
OUT_DIR="${OUT_DIR:-/home/nikhileswara/DEIMv2/outputs/lightly_distill/dinov3_vits16plus_from_vitb16_crowd_human_mot}"
STUDENT_MODEL="${STUDENT_MODEL:-dinov3/vits16plus}"
TEACHER_MODEL="${TEACHER_MODEL:-dinov3/vitb16}"
METHOD="${METHOD:-distillationv2}"
EPOCHS="${EPOCHS:-200}"
BATCH_SIZE="${BATCH_SIZE:-512}"
NUM_WORKERS="${NUM_WORKERS:-16}"
DEVICES="${DEVICES:-4}"
PRECISION="${PRECISION:-32-true}"
SEED="${SEED:-143}"
OVERWRITE="${OVERWRITE:-0}"
RESUME_INTERRUPTED="${RESUME_INTERRUPTED:-0}"

RUNNER="${ROOT_DIR}/engine/lightly_distill_deimv2/run_distill_dinov3_splus_vitb.py"
LOG_DIR="${LOG_DIR:-${ROOT_DIR}/logs/lightly_distill}"
LOG_FILE="${LOG_FILE:-${LOG_DIR}/$(basename "${OUT_DIR}").log}"

if [[ ! -f "${RUNNER}" ]]; then
  echo "Runner not found: ${RUNNER}" >&2
  exit 1
fi

if [[ ! -d "${DATA_DIR}" ]]; then
  echo "Data directory not found: ${DATA_DIR}" >&2
  exit 1
fi

# Backward-compat cleanup: old launcher wrote live.log inside OUT_DIR, which
# makes Lightly refuse to start because the output dir is not empty.
if [[ -d "${OUT_DIR}" && -f "${OUT_DIR}/live.log" ]]; then
  if [[ -z "$(find "${OUT_DIR}" -mindepth 1 -maxdepth 1 ! -name 'live.log' -print -quit)" ]]; then
    rm -f "${OUT_DIR}/live.log"
    rmdir "${OUT_DIR}" 2>/dev/null || true
  fi
fi

mkdir -p "$(dirname "${LOG_FILE}")"

echo "Starting LightlyTrain distillation"
echo "CUDA_VISIBLE_DEVICES=${CUDA_DEVICES}"
echo "DATA_DIR=${DATA_DIR}"
echo "OUT_DIR=${OUT_DIR}"
echo "LOG_FILE=${LOG_FILE}"

CMD=(
  conda run --no-capture-output -n "${CONDA_ENV}" python -u "${RUNNER}"
  --data "${DATA_DIR}"
  --out "${OUT_DIR}"
  --student-model "${STUDENT_MODEL}"
  --teacher-model "${TEACHER_MODEL}"
  --method "${METHOD}"
  --epochs "${EPOCHS}"
  --batch-size "${BATCH_SIZE}"
  --num-workers "${NUM_WORKERS}"
  --devices "${DEVICES}"
  --precision "${PRECISION}"
  --seed "${SEED}"
)

if [[ "${OVERWRITE}" == "1" ]]; then
  CMD+=(--overwrite)
fi
if [[ "${RESUME_INTERRUPTED}" == "1" ]]; then
  CMD+=(--resume-interrupted)
fi

CUDA_VISIBLE_DEVICES="${CUDA_DEVICES}" \
PYTHONUNBUFFERED=1 \
  "${CMD[@]}" 2>&1 | tee "${LOG_FILE}"