Complete standalone final deliverable and unaligned Q2 results

2026-09-25 22:22:37 +08:00
parent c6b018e5d0
commit adc9c2064b
267 changed files with 15479 additions and 7976 deletions
@@ -0,0 +1,51 @@
+"""Read the native Q1 artifacts needed by the unified alignment adapter."""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from typing import Any
+
+import numpy as np
+
+FEATURE_DIR = Path(__file__).resolve().parents[1] / "output" / "q1" / "features_v2"
+
+
+def load_sample(sample_id: str, feature_dir: Path = FEATURE_DIR) -> dict[str, Any]:
+    """Load a Q1 sample by its video/clip ID from a feature directory."""
+    feature_dir = Path(feature_dir)
+    manifest = feature_dir / "manifest_q1.jsonl"
+    for line in manifest.read_text(encoding="utf-8").splitlines():
+        row = json.loads(line)
+        if row["sample_id"] != sample_id:
+            continue
+        path = feature_dir / Path(row["feature_path"]).name
+        with np.load(path, allow_pickle=False) as archive:
+            result = {name: archive[name] for name in archive.files}
+        result["_path"] = path
+        result["_manifest"] = row
+        result["_meta"] = json.loads(str(result["meta_json"]))
+        return result
+    raise KeyError(f"sample_id not found: {sample_id}")
+
+
+def native_arrays(sample: dict[str, Any], modality: str):
+    """Return values, observed dimensions, time intervals and quality evidence."""
+    if modality == "text":
+        values = sample["native_text_features"].astype(np.float32)
+        return (
+            values,
+            np.broadcast_to(sample["native_text_observed"][:, None], values.shape),
+            sample["native_text_intervals"].astype(np.float32),
+            sample["native_text_quality_effective"].astype(np.float32),
+            sample["native_text_quality_available"].astype(bool),
+        )
+    if modality in ("audio", "vision"):
+        prefix = f"native_{modality}_"
+        return (
+            sample[prefix + "features"].astype(np.float32),
+            sample[prefix + "mask"].astype(bool),
+            sample[prefix + "intervals"].astype(np.float32),
+            sample[prefix + "quality"].astype(np.float32),
+            sample[prefix + "quality_available"].astype(bool),
+        )
+    raise ValueError(f"unknown modality: {modality}")