Datasets:
File size: 6,901 Bytes
aa991fc 358e603 aa991fc 358e603 aa991fc 358e603 aa991fc | 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 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 | #!/usr/bin/env python
"""Assemble one capture into the HuggingFace staging tree.
staging/data/<PxCy>/videos/cam00.mp4 .. cam31.mp4 hardlink to frameset-videos/NN.mp4
staging/data/<PxCy>/cameras.json copy of the capture calibration
staging/data/<PxCy>/smplx.npz written by consolidate_smplx.py
staging/data/<PxCy>/preview.jpg written by make_previews.py
staging/data/<PxCy>/capture.json machine-readable capture card
Videos are HARDLINKED (same filesystem, /mnt/sdb) so staging costs ~0 bytes and the
source capture dir is never modified. Falls back to symlink, then copy.
The source `frameset-videos/` must already be extracted (`7z x frameset-videos.7z`);
--extract will do it if the directory is missing.
Usage:
python stage_capture.py P1C1 --staging /mnt/sdb/degas_project/DREAMS-AVATAR-hf/staging
"""
from __future__ import annotations
import argparse
import json
import os
import shutil
import subprocess
import sys
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parent))
from load_capture import VIDEO_FRAME_OFFSET # noqa: E402 single source of truth for d
DATASET_ROOT = Path("/mnt/sdb/degas_project/DEGAS_DATASET")
# The published set is 10 captures. C1 = train, C2 = test.
# P1..P4 ship both sessions; P5/P6 are CROSS-REENACT DRIVING ONLY, so only their C2 is
# published. P5C1 does not exist on disk at all; P6C1 exists but is deliberately excluded.
INCLUDED = ["P1C1", "P1C2", "P2C1", "P2C2", "P3C1", "P3C2",
"P4C1", "P4C2", "P5C2", "P6C2"]
EXCLUDED = {
"P5C1": "does not exist",
"P6C1": "excluded on purpose: P5/P6 contribute cross-reenact driving (C2) only",
}
def link_or_copy(src: Path, dst: Path) -> str:
if dst.exists() or dst.is_symlink():
dst.unlink()
dst.parent.mkdir(parents=True, exist_ok=True)
try:
os.link(src, dst)
return "hardlink"
except OSError:
try:
dst.symlink_to(src)
return "symlink"
except OSError:
shutil.copy2(src, dst)
return "copy"
def probe_video(path: Path) -> dict:
out = subprocess.run(
["ffprobe", "-v", "error", "-select_streams", "v:0", "-show_entries",
"stream=width,height,nb_frames,r_frame_rate,codec_name,pix_fmt",
"-of", "json", str(path)],
capture_output=True, text=True, check=True).stdout
s = json.loads(out)["streams"][0]
num, den = s["r_frame_rate"].split("/")
return {"width": int(s["width"]), "height": int(s["height"]),
"n_frames": int(s["nb_frames"]), "fps": round(int(num) / int(den), 4),
"codec": s["codec_name"], "pix_fmt": s["pix_fmt"]}
def stage(capture: str, dataset_root: Path, staging: Path, extract: bool = False,
force: bool = False) -> dict:
if capture in EXCLUDED and not force:
raise SystemExit(
f"{capture} is NOT part of DREAMS-AVATAR: {EXCLUDED[capture]}.\n"
f"published captures: {', '.join(INCLUDED)} (--force to override)")
if capture not in INCLUDED and not force:
raise SystemExit(f"{capture} is not in the published set "
f"({', '.join(INCLUDED)}); use --force to add it anyway")
src = dataset_root / capture
if not src.is_dir():
raise SystemExit(f"no such capture: {src}")
vdir = src / "frameset-videos"
if not vdir.is_dir() or not list(vdir.glob("*.mp4")):
if not extract:
raise SystemExit(f"{vdir} empty; re-run with --extract (runs 7z x)")
arc = src / "frameset-videos.7z"
print(f"[stage] extracting {arc} ...", flush=True)
subprocess.run(["7z", "x", "-y", f"-o{src}", str(arc)], check=True,
stdout=subprocess.DEVNULL)
mp4s = sorted(vdir.glob("*.mp4"))
dst = staging / "data" / capture
mode = None
total = 0
for m in mp4s:
cam = int(m.stem)
mode = link_or_copy(m, dst / "videos" / f"cam{cam:02d}.mp4")
total += m.stat().st_size
print(f"[stage] {capture}: {len(mp4s)} videos via {mode} "
f"({total / 2**30:.2f} GiB)", flush=True)
cams_src = src / "cameras.json"
shutil.copy2(cams_src, dst / "cameras.json")
cams = json.loads(cams_src.read_text())
n_rigs = len(cams["rigs"])
v0 = probe_video(mp4s[0])
assert v0["width"] % 2 == 0
card = {
"capture": capture,
"subject": capture[:2],
"session": capture[2:],
"role": ("train" if capture.endswith("C1") else
("cross_reenact_driving" if capture[:2] in ("P5", "P6") else "test")),
"n_cams": len(mp4s),
"n_rigs_in_calibration": n_rigs,
"video": {**v0,
"layout": "side-by-side: LEFT half = RGB, RIGHT half = alpha matte",
"rgb_width": v0["width"] // 2},
"video_bytes": total,
"frame_convention": {
"offset_d": VIDEO_FRAME_OFFSET,
"rule": "video frame index == smplx frame index == GT frame index (d=0)",
"smplx_frame": "0-based frame id, stored in smplx.npz['frames']",
"video_frame": "the same number, 0-based index into camNN.mp4",
"measured": "d is measured per capture by verify_alignment.py, not assumed",
},
"camera_convention": {
"K": "[[fx,0,cx],[0,fy,cy],[0,0,1]] from rigs[i].cameras[0]",
"R_w2c": "R @ diag(1,-1,-1)",
"t_w2c": "-R @ c",
"note": "cameras.json world is Y-down; the world_flip puts the fit in the "
"same Y-up world as the DEGAS GT",
},
}
smplx_npz = dst / "smplx.npz"
if smplx_npz.exists():
import numpy as np
z = np.load(smplx_npz, allow_pickle=False)
card["n_frames"] = int(len(z["frames"]))
card["frame_range"] = [int(z["frames"][0]), int(z["frames"][-1])]
card["smplx_kwargs"] = json.loads(str(z["smplx_kwargs"]))
(dst / "capture.json").write_text(json.dumps(card, indent=2) + "\n")
print(f"[stage] wrote {dst / 'capture.json'}", flush=True)
return card
def main() -> int:
ap = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("capture")
ap.add_argument("--dataset-root", type=Path, default=DATASET_ROOT)
ap.add_argument("--staging", type=Path, required=True)
ap.add_argument("--extract", action="store_true", help="7z x frameset-videos.7z if missing")
ap.add_argument("--force", action="store_true",
help="stage a capture outside the published 10 (P5C1/P6C1 are excluded)")
a = ap.parse_args()
print(json.dumps(stage(a.capture, a.dataset_root, a.staging, a.extract, a.force)))
return 0
if __name__ == "__main__":
sys.exit(main())
|