Datasets:
File size: 3,715 Bytes
96e7a80 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 | #!/usr/bin/env python3
"""Export RealEstate10K-Cap to per-clip .npy files in the TKCAM directory layout.
<out_dir>/
trajectories/<id>.npy (N, 12) TKCAM format, or (N, 9) universal format; row i = frame i+1
texts/<id>.txt "caption#word/POS ..." (TKCAM text loader format)
untagged_text/<id>.txt plain caption
train.txt / val.txt / test.txt
Mean.npy / Std.npy (12d only; over all exported clips, as in TKCAM)
Formats:
12d [x, y, z, dx, dy, dz, r11, r21, r31, r12, r22, r32] (as stored)
9d [r11, r21, r31, r12, r22, r32, x, y, z] (CLaTr / benchmark format)
Requirements: pip install numpy pyarrow
Example:
python scripts/export_npy.py --data-dir data --out-dir RealEstate10K-Cap_npy --format 12d
# frames for TKCAM's --frame_dir:
python scripts/download_frames.py --data-dir data --out-dir RealEstate10K-Cap_frames
"""
import argparse
import glob
import os
import shutil
import numpy as np
import pyarrow.parquet as pq
def to_9d(traj12):
return np.concatenate([traj12[:, 6:12], traj12[:, 0:3]], axis=1)
def main():
parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument("--data-dir", default="data", help="folder with the *.parquet files")
parser.add_argument("--out-dir", required=True)
parser.add_argument("--format", choices=["12d", "9d"], default="12d")
parser.add_argument("--exclude-benchmark", action="store_true",
help="drop clips with in_benchmark_test=True from train/val")
args = parser.parse_args()
for sub in ("trajectories", "texts", "untagged_text"):
os.makedirs(os.path.join(args.out_dir, sub), exist_ok=True)
cols = ["id", "split", "caption", "caption_pos", "trajectory_12d", "in_benchmark_test"]
splits = {"train": [], "val": [], "test": []}
for f in sorted(glob.glob(os.path.join(args.data_dir, "*.parquet"))):
for row in pq.read_table(f, columns=cols).to_pylist():
if args.exclude_benchmark and row["in_benchmark_test"] and row["split"] != "test":
continue
sid = row["id"]
traj = np.asarray(row["trajectory_12d"], dtype=np.float32)
np.save(os.path.join(args.out_dir, "trajectories", f"{sid}.npy"),
to_9d(traj) if args.format == "9d" else traj)
with open(os.path.join(args.out_dir, "texts", f"{sid}.txt"), "w") as fh:
fh.write(f"{row['caption']}#{row['caption_pos']}\n")
with open(os.path.join(args.out_dir, "untagged_text", f"{sid}.txt"), "w") as fh:
fh.write(row["caption"] + "\n")
splits[row["split"]].append(sid)
for split, ids in splits.items():
with open(os.path.join(args.out_dir, f"{split}.txt"), "w") as fh:
fh.write("\n".join(sorted(ids)) + "\n")
if args.format == "12d":
stats_dir = os.path.join(os.path.dirname(os.path.abspath(args.data_dir)), "stats")
if not args.exclude_benchmark and os.path.isdir(stats_dir):
for name in ("Mean.npy", "Std.npy"):
shutil.copy2(os.path.join(stats_dir, name), os.path.join(args.out_dir, name))
else:
frames = np.concatenate([np.load(os.path.join(args.out_dir, "trajectories", f"{s}.npy"))
for ids in splits.values() for s in ids])
np.save(os.path.join(args.out_dir, "Mean.npy"), frames.mean(0))
np.save(os.path.join(args.out_dir, "Std.npy"), frames.std(0))
print({k: len(v) for k, v in splits.items()}, "->", args.out_dir)
if __name__ == "__main__":
main()
|