language_table/download.sh

#!/bin/bash

# Usage: ./download.sh [num_trajectories]
# Example: ./download.sh 5000
# Use -1 to download all shards.

NUM_TRAJS=${1:-5000}
DEST_DIR=$(dirname "$(readlink -f "$0")")

echo "Target directory: $DEST_DIR"

# Download metadata
echo "Downloading metadata files..."
gsutil cp gs://gresearch/robotics/language_table/0.1.0/dataset_info.json "$DEST_DIR/"
gsutil cp gs://gresearch/robotics/language_table/0.1.0/features.json "$DEST_DIR/"

if [ "$NUM_TRAJS" -eq -1 ]; then
    echo "Downloading all 1024 shards (approx 400GB)..."
    gsutil -m cp "gs://gresearch/robotics/language_table/0.1.0/language_table-train.tfrecord-*-of-01024" "$DEST_DIR/"
else
    # Each shard has ~430 trajectories. 
    NUM_SHARDS=$(( (NUM_TRAJS + 399) / 400 ))
    
    if [ "$NUM_SHARDS" -gt 1024 ]; then
        NUM_SHARDS=1024
    fi

    echo "Downloading $NUM_SHARDS shards to get approximately $NUM_TRAJS trajectories..."
    
    # Loop to download shards
    for i in $(seq 0 $((NUM_SHARDS-1))); do
        SHARD_ID=$(printf "%05d" $i)
        FILE="language_table-train.tfrecord-$SHARD_ID-of-01024"
        if [ ! -f "$DEST_DIR/$FILE" ]; then
            echo "Downloading $FILE..."
            gsutil cp "gs://gresearch/robotics/language_table/0.1.0/$FILE" "$DEST_DIR/"
        else
            echo "$FILE already exists, skipping."
        fi
    done
fi

echo "Download process finished."