Upload scripts/validate_clean_metadata.py

scripts/validate_clean_metadata.py

@@ -0,0 +1,46 @@
+
+import csv
+import os
+import time
+from collections import Counter
+from pathlib import Path
+
+root = Path('/aifs4su/yezhen/data/Talker-T2AV-Data-clean')
+csv_path = root / 'metadata' / 'train.csv'
+report_path = root / 'validation_report.txt'
+cols = ['wav_path', 'motion_pt_path', 'video_path']
+rows = 0
+missing = Counter()
+sources = Counter()
+start = time.time()
+examples = []
+old_name_rows = 0
+
+with csv_path.open('r', newline='', encoding='utf-8') as f:
+    reader = csv.DictReader(f)
+    for row in reader:
+        rows += 1
+        sources[row.get('dataset_source', '')] += 1
+        if any('seamless_emilia' in row.get(k, '') for k in row.keys()):
+            old_name_rows += 1
+        for col in cols:
+            rel = row.get(col, '')
+            if rel and not (root / rel).exists():
+                missing[col] += 1
+                if len(examples) < 20:
+                    examples.append((rows, col, rel))
+        if rows % 10000 == 0:
+            print(f'validated={rows} elapsed={time.time()-start:.1f}s missing={dict(missing)} old_name_rows={old_name_rows}', flush=True)
+
+elapsed = time.time() - start
+with report_path.open('w', encoding='utf-8') as out:
+    out.write(f'rows={rows}\n')
+    out.write(f'sources={dict(sources)}\n')
+    out.write(f'missing={dict(missing)}\n')
+    out.write(f'old_name_rows={old_name_rows}\n')
+    out.write(f'elapsed_sec={elapsed:.1f}\n')
+    for ex in examples:
+        out.write(f'missing_example={ex}\n')
+print(f'DONE rows={rows} missing={dict(missing)} old_name_rows={old_name_rows} elapsed={elapsed:.1f}s report={report_path}', flush=True)
+if missing or old_name_rows:
+    raise SystemExit(2)