PULSE-code / tests /conftest.py
velvet-pine-22's picture
PULSE-code: benchmark code for the PULSE dataset (camera-ready release)
dffbc1f verified
Raw History Blame Contribute Delete
7.09 kB
"""Synthetic miniature PULSE tree with the public layout and its edge cases."""
from __future__ import annotations
import csv
import json
import os
import sys
from pathlib import Path
import numpy as np
import pytest
REPO = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(REPO))
from pulse.data import modalities as M # noqa: E402
T = 240 # frames per synthetic recording (100 Hz)
IMU_FIELDS = ["acc_x", "acc_y", "acc_z", "gyr_x", "gyr_y", "gyr_z", "ang_x", "ang_y", "ang_z",
"mag_x", "mag_y", "mag_z", "q0", "q1", "q2", "q3", "temp", "battery"]
SKELETON_JOINTS = ["Hips", "Spine", "RightHand", "RightInHandIndex", "LeftHand"]
def imu_columns(n_sensors: int):
return [f"WT{s}_{f}" for s in range(n_sensors) for f in IMU_FIELDS]
def write_csv(path: Path, header, data, extra_cols=None):
path.parent.mkdir(parents=True, exist_ok=True)
with open(path, "w", newline="", encoding="utf-8") as f:
w = csv.writer(f)
w.writerow(["time"] + list(header))
for i, row in enumerate(data):
w.writerow([f"{i / 100:.2f}"] + [f"{x:.5f}" if isinstance(x, float) else x for x in row])
def write_tsv(path: Path, n: int, rng):
path.parent.mkdir(parents=True, exist_ok=True)
cols = ["Frame", "Time"]
for j in SKELETON_JOINTS:
cols += [f"{j}_X", f"{j}_Y", f"{j}_Z", f"{j}_QX", f"{j}_QY", f"{j}_QZ", f"{j}_QW"]
with open(path, "w", encoding="utf-8") as f:
f.write("\t".join(cols) + "\n")
for i in range(n):
vals = [str(i), f"{i / 100:.2f}"] + [f"{v:.4f}" for v in rng.normal(500, 50, len(cols) - 2)]
f.write("\t".join(vals) + "\n")
# volunteer -> scenario -> spec
LAYOUT = {
"v1": {"s1": {}, "s2": {"tsv_at_root": True}, "s9": {"no_annotation": True}},
"v2": {"s1": {}, "s2": {"eyetrack_one_eye": True}},
"v3": {"s1": {"no_pressure": True}, "s2": {"no_pressure": True}},
"v5": {"s1": {"imu_sensors": 4, "no_mocap": True}, "s2": {"imu_sensors": 4, "no_mocap": True}},
"v6": {"s1": {}, "s2": {}},
}
def build_tree(root: Path, seed: int = 0) -> Path:
rng = np.random.default_rng(seed)
rows = []
for vol, scens in LAYOUT.items():
for scen, spec in scens.items():
rid = f"{vol}{scen}"
d = root / "data" / vol / scen
d.mkdir(parents=True, exist_ok=True)
mods = ["emg", "imu"]
k = int(scen[1:])
write_csv(d / "aligned_emg_100hz.csv", M.EMG_COLS,
(rng.normal(k, 1, (T, 8))).tolist())
write_csv(d / "aligned_imu_100hz.csv", imu_columns(spec.get("imu_sensors", 10)),
rng.normal(k, 1, (T, 18 * spec.get("imu_sensors", 10))).tolist())
write_csv(d / "aligned_myo_quat_100hz.csv", M.MYO_QUAT_COLS,
rng.normal(0, 1, (T, 4)).tolist())
ecols = list(M.EYETRACK_CORE_COLS)
if spec.get("eyetrack_one_eye"):
ecols = [c for c in ecols if "Right Eye" not in c]
with open(d / "aligned_eyetrack_100hz.csv", "w", newline="", encoding="utf-8") as f:
w = csv.writer(f)
w.writerow(["time", "UTC"] + ecols)
for i in range(T):
w.writerow([i / 100, 1.7e12 + i] + list(rng.normal(100, 10, len(ecols))))
mods.append("eyetrack")
if not spec.get("no_pressure"):
p = np.abs(rng.normal(0, 1, (T, 50)))
p[: T // 2] = 0.0
write_csv(d / "aligned_pressure_100hz.csv", M.PRESSURE_COLS, p.tolist())
mods.append("pressure")
if not spec.get("no_mocap"):
tsv = (d if spec.get("tsv_at_root") else d / "raw") / f"aligned_{rid}_s_Q.tsv"
write_tsv(tsv, T, rng)
write_csv(d / "aligned_mocap_100hz.csv", ["Q_RThumbTip X", "Q_RThumbTip Y",
"Q_RThumbTip Z", "Q_RThumbTip Type"],
[[*rng.normal(0, 1, 3).tolist(), "Measured"] for _ in range(T)])
mods.append("mocap")
align = ["emg", "imu", "eyetrack", "myo_quat"]
align += ["pressure"] if "pressure" in mods else []
align += ["mocap", "mocap_sq"] if "mocap" in mods else []
with open(d / "alignment_metadata.json", "w", encoding="utf-8") as f:
json.dump({"volunteer": vol, "scenario": scen, "n_samples": T,
"aligned_modalities": align, "modalities": {m: {} for m in align}}, f)
has_ann = scen != "s9" and not spec.get("no_annotation")
if has_ann:
segs = []
verbs = ["grasp", "move", "place", "support", "grasp"]
nouns = ["cup", "box", "cup", "cup", "mouse"]
for i in range(5):
segs.append({"timestamp": f"00:00:0{i}-00:00:0{i + 1}",
"action_annotation": {"action_name": verbs[i], "object_name": nouns[i],
"hand_type": "right"}})
(root / "annotations" / vol).mkdir(parents=True, exist_ok=True)
with open(root / "annotations" / vol / f"{scen}.json", "w", encoding="utf-8") as f:
json.dump({"segments": segs}, f)
all_mods = ["scene_video", "eyetrack", "pressure", "emg", "imu", "mocap"]
rows.append({"recording_id": rid, "volunteer_id": vol, "scenario_id": scen,
"duration_sec": T / 100, "n_samples_100hz": T,
"aligned_modalities": "|".join(m for m in all_mods if m in mods),
"missing_modalities": "|".join(m for m in all_mods if m not in mods
and m != "scene_video"),
"has_dense_annotations": int(has_ann), "split": "train"})
(root / "metadata").mkdir(parents=True, exist_ok=True)
with open(root / "metadata" / "recordings.csv", "w", newline="", encoding="utf-8") as f:
w = csv.DictWriter(f, fieldnames=list(rows[0]))
w.writeheader()
w.writerows(rows)
return root
@pytest.fixture()
def mini_root(tmp_path) -> Path:
return build_tree(tmp_path / "PULSE")
def mini_split(path: Path) -> Path:
"""A 3-fold split over the synthetic volunteers."""
d = {"schema": "pulse-split/1", "name": "mini", "volunteers": ["v1", "v2", "v3", "v5", "v6"],
"folds": [
{"fold": 0, "test": ["v1"], "inner_val": ["v2"], "train": ["v3", "v5", "v6"]},
{"fold": 1, "test": ["v2", "v5"], "inner_val": ["v6"], "train": ["v1", "v3"]},
{"fold": 2, "test": ["v3", "v6"], "inner_val": ["v1"], "train": ["v2", "v5"]},
]}
path.write_text(json.dumps(d), encoding="utf-8")
return path
def real_root():
r = os.environ.get("PULSE_ROOT")
return Path(r) if r and Path(r, "data").is_dir() else None
realdata = pytest.mark.skipif(real_root() is None, reason="PULSE_ROOT not set to a dataset copy")