train.sh

set -e
SCENE_IDS="1d003b07bd/dslr 1f7cbbdde1/dslr"
DATA_ROOT="data/scnt/datasets/"

STEP=0.002
FOVMOD_TRAIN=1.3

ITERS_NUM=30000

for SCENE_ID in $SCENE_IDS; do
    echo "Processing scene: $SCENE_ID"

    DATASET_DIR="$DATA_ROOT$SCENE_ID/"
    OUTPUT_DIR="./output/scnt/$SCENE_ID"

    BEAP_DIR_TRAIN="beap_fov_${FOVMOD_TRAIN}_step_${STEP}/"

    TRAIN_MASK_FN="fov_${FOVMOD_TRAIN}_step_${STEP}_mask.png"

    # Train: Generate beap
    python data/scnt/scnt_kb2beap.py --path "$DATASET_DIR" --dst "$BEAP_DIR_TRAIN" --step "$STEP" --fov_mod "$FOVMOD_TRAIN" --mask_dst "$TRAIN_MASK_FN"

    # Train: Run training script
    python train.py -s "$DATASET_DIR" -m "$OUTPUT_DIR" \
        --iterations "$ITERS_NUM" \
        --checkpoint_iterations 3000 7000 15000 30000 \
        --save_iterations 3000 7000 15000 30000 \
        --test_iterations 3000 7000 15000 30000 \
        --resolution 1 \
        --eval \
        --sample_step "$STEP" --fov_mod "$FOVMOD_TRAIN" \
        --mask_path "${DATASET_DIR}${BEAP_DIR_TRAIN}${TRAIN_MASK_FN}" \
        --sibr_mask_refcam "${DATASET_DIR}colmap/cameras_fish.txt"
done