Add extract_ego4d_batch.py for efficient EgoSpeak clip extraction. This script loads the Qwen model once and processes selected manifest rows sequentially, ensuring each clip is handled independently. Update README with usage instructions and details on the extraction process, including handling of completed outputs and memory management.

streaming_turn_model_directly.py

@@ -59,7 +59,7 @@ import subprocess
 from collections import OrderedDict
 from dataclasses import asdict, dataclass
 from pathlib import Path
-from typing import Any
+from typing import Any, Sequence
 
 import numpy as np
 import torch
@@ -778,7 +778,7 @@ def _parse_crop(value: str | None) -> tuple[int, int, int, int] | None:
     return values
 
 
-def parse_args() -> argparse.Namespace:
+def parse_args(argv: Sequence[str] | None = None) -> argparse.Namespace:
     parser = argparse.ArgumentParser(description=__doc__)
     source = parser.add_mutually_exclusive_group(required=True)
     source.add_argument("--video", type=Path, help="One original video clip")
@@ -873,7 +873,7 @@ def parse_args() -> argparse.Namespace:
         action="store_true",
         help="Validate/index direct media and exit before loading Qwen",
     )
-    return parser.parse_args()
+    return parser.parse_args(argv)
 
 
 def _resolve_duration_and_target(
@@ -941,9 +941,23 @@ def _print_media_summary(reader: CausalSourceAV, start: float, end: float) -> No
         )
 
 
-def main() -> None:
-    args = parse_args()
+def main(
+    argv: Sequence[str] | None = None,
+    *,
+    model: Any | None = None,
+    processor: Any | None = None,
+) -> None:
+    """Run one direct extraction, optionally reusing loaded Qwen objects.
+
+    ``model`` and ``processor`` must be supplied together.  Reusing them is
+    safe across independent clips because this function constructs a new media
+    reader, extractor, and temporal cache on every call.
+    """
+
+    args = parse_args(argv)
     core._configure_qwen_logging()
+    if (model is None) != (processor is None):
+        raise ValueError("model and processor must be supplied together")
     if args.verify_every < 0 or args.verify_continuous_every < 0:
         raise SystemExit("verification intervals must be >= 0")
     if args.frame_cache_size < 0:
@@ -1055,12 +1069,15 @@ def main() -> None:
     )
     config.validate()
 
-    model, processor = core.load_qwen_omni(
-        args.model_id,
-        dtype=args.dtype,
-        device_map={"": args.device},
-        use_flash_attention_2=args.flash_attention_2,
-    )
+    if model is None:
+        model, processor = core.load_qwen_omni(
+            args.model_id,
+            dtype=args.dtype,
+            device_map={"": args.device},
+            use_flash_attention_2=args.flash_attention_2,
+        )
+    else:
+        model.eval()
     model.requires_grad_(False)
     media_path_for_api = reader.clips[0].path