Add bulk frame ingest for recording sessions

Registering thousands of extracted frames one-by-one via POST /frames
doesn't scale, so add `spelunkai-labeling-backend ingest-session`: it
writes Frame rows directly against the database (no server needs to
be running) for all frame_*.png files under a directory already
placed under FRAMES_ROOT. Idempotent per (session_name, frame_index),
so re-running after copying more frames only inserts the new ones.

Restructured the CLI into subcommands (serve / ingest-session) while
keeping `spelunkai-labeling-backend` with no arguments working exactly
as before (defaults to serve), verified against a live run. 22/22
backend tests pass.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

labeling/backend/README.md

@@ -4,11 +4,13 @@ API + data model for the bounding-box labeling tool. See CLAUDE.md §3.2 for the
 full requirements this is built against.
 
 **Status:** core data model + CRUD API implemented (sets, hierarchical classes,
-frames, labels, per-frame/per-set status). **Not yet implemented:** dataset
-versioning/promotion (`enemy-v1`, `enemy-v2`, ...) — deferred since it needs its
-own design pass (snapshot semantics: labels are editable at any time, but a
-promoted dataset version must stay reproducible). Also not yet implemented: any
-frontend, auth/login (see below), or the active-learning auto-label workflow.
+frames, labels, per-frame/per-set status), frame image serving, and bulk frame
+ingest from a recording session (see `labeling/frontend/` for the UI). **Not yet
+implemented:** dataset versioning/promotion (`enemy-v1`, `enemy-v2`, ...) —
+deferred since it needs its own design pass (snapshot semantics: labels are
+editable at any time, but a promoted dataset version must stay reproducible).
+Also not yet implemented: auth/login (see below) or the active-learning
+auto-label workflow.
 
 ## Stack
 
@@ -31,6 +33,9 @@ frontend, auth/login (see below), or the active-learning auto-label workflow.
 - `FrameSetStatus` — per-frame, per-set label state (`unlabeled` / `auto_labeled` /
   `reviewed`).
 
+Frame images are served from `FRAMES_ROOT` (env var, default `./frames`) at
+`/images/<image_path>`; `Frame.image_path` is always relative to that root.
+
 ## Setup
 
 ```
@@ -48,6 +53,24 @@ spelunkai-labeling-backend
 
 Interactive API docs at `http://127.0.0.1:8000/docs` once running.
 
+## Bulk-ingesting a recording session
+
+After a session's extracted frames (`recording/`'s `extract-frames` output) have
+been copied/rsynced onto this machine under `FRAMES_ROOT`, register them all in
+one shot (writes directly to the database, no server needs to be running):
+
+```
+spelunkai-labeling-backend ingest-session \
+  --frames-dir /path/under/FRAMES_ROOT/run01_frames \
+  --session-name run01
+```
+
+`--image-path-prefix` defaults to the frames directory's own name (here
+`run01_frames`) — override it if the directory was copied under a different name.
+Safe to re-run: frames already registered for that session (by frame index) are
+skipped. `--width`/`--height` default to 1280x720 (Spelunky Classic HD's fixed
+capture resolution).
+
 ## Testing
 
 ```

labeling/backend/src/spelunkai_labeling_backend/cli.py

@@ -1,13 +1,73 @@
-"""Convenience entry point to run the labeling backend with uvicorn."""
+"""Command-line entry points for the labeling backend: run the server, or
+bulk-ingest a recording session's extracted frames.
+"""
 from __future__ import annotations
 
+import argparse
+import os
+from pathlib import Path
+
 import uvicorn
 
+from .db import Database
+from .ingest import ingest_session
 from .main import create_app
 
 
 def main() -> None:
-    uvicorn.run(create_app(), host="127.0.0.1", port=8000)
+    parser = argparse.ArgumentParser(prog="spelunkai-labeling-backend")
+    parser.set_defaults(command="serve", host="127.0.0.1", port=8000)
+    subparsers = parser.add_subparsers(dest="command")
+
+    serve = subparsers.add_parser("serve", help="Run the API server (default if no subcommand is given)")
+    serve.add_argument("--host", default="127.0.0.1")
+    serve.add_argument("--port", type=int, default=8000)
+
+    ingest = subparsers.add_parser(
+        "ingest-session",
+        help="Bulk-register a recording session's extracted frames (must already be under FRAMES_ROOT)",
+    )
+    ingest.add_argument(
+        "--frames-dir", type=Path, required=True,
+        help="Directory of frame_*.png files, already placed under FRAMES_ROOT",
+    )
+    ingest.add_argument("--session-name", required=True)
+    ingest.add_argument(
+        "--image-path-prefix", default=None,
+        help="Path stored on each frame, relative to FRAMES_ROOT (default: --frames-dir's own name)",
+    )
+    ingest.add_argument("--width", type=int, default=1280)
+    ingest.add_argument("--height", type=int, default=720)
+
+    args = parser.parse_args()
+
+    if args.command == "ingest-session":
+        _run_ingest(args)
+    else:
+        _run_serve(args)
+
+
+def _run_serve(args: argparse.Namespace) -> None:
+    uvicorn.run(create_app(), host=args.host, port=args.port)
+
+
+def _run_ingest(args: argparse.Namespace) -> None:
+    database = Database(os.environ.get("DATABASE_URL"))
+    database.init_models()
+
+    prefix = args.image_path_prefix or args.frames_dir.name
+    result = ingest_session(
+        database=database,
+        frames_dir=args.frames_dir,
+        session_name=args.session_name,
+        image_path_prefix=prefix,
+        width=args.width,
+        height=args.height,
+    )
+    print(
+        f"Ingested {result.inserted} new frame(s) for session '{args.session_name}' "
+        f"({result.already_existed} already existed, {result.total_seen} total seen)."
+    )
 
 
 if __name__ == "__main__":

labeling/backend/src/spelunkai_labeling_backend/ingest.py

@@ -0,0 +1,77 @@
+"""Bulk-register a recording session's extracted frames as `Frame` rows.
+
+Meant to run on the same machine as the labeling backend's database (`jai`),
+after a session's frame images have been copied/rsynced under `FRAMES_ROOT` -
+inserts directly against the database rather than over HTTP, since a session
+can have thousands of frames and this is a local, same-machine operation.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Iterator, List
+
+from . import models
+from .db import Database
+
+FRAME_GLOB = "frame_*.png"
+
+
+@dataclass
+class IngestResult:
+    inserted: int
+    already_existed: int
+    total_seen: int
+
+
+def iter_frame_files(frames_dir: Path) -> Iterator[Path]:
+    return sorted(frames_dir.glob(FRAME_GLOB))
+
+
+def frame_index_from_filename(path: Path) -> int:
+    # "frame_000123.png" -> 123, matching recording/spelunkai_recording/frames.py's naming.
+    return int(path.stem.split("_")[-1])
+
+
+def ingest_session(
+    database: Database,
+    frames_dir: Path,
+    session_name: str,
+    image_path_prefix: str,
+    width: int,
+    height: int,
+) -> IngestResult:
+    frame_files = list(iter_frame_files(frames_dir))
+    if not frame_files:
+        raise FileNotFoundError(f"no '{FRAME_GLOB}' files found in {frames_dir}")
+
+    db_session = database.session_factory()
+    try:
+        existing_indices = {
+            row.frame_index
+            for row in db_session.query(models.Frame.frame_index).filter_by(session_name=session_name)
+        }
+
+        new_frames: List[models.Frame] = []
+        for frame_file in frame_files:
+            frame_index = frame_index_from_filename(frame_file)
+            if frame_index in existing_indices:
+                continue
+            new_frames.append(models.Frame(
+                session_name=session_name,
+                frame_index=frame_index,
+                image_path=f"{image_path_prefix}/{frame_file.name}",
+                width=width,
+                height=height,
+            ))
+
+        db_session.add_all(new_frames)
+        db_session.commit()
+    finally:
+        db_session.close()
+
+    return IngestResult(
+        inserted=len(new_frames),
+        already_existed=len(frame_files) - len(new_frames),
+        total_seen=len(frame_files),
+    )

labeling/backend/tests/test_ingest.py

@@ -0,0 +1,88 @@
+from pathlib import Path
+
+import pytest
+
+from spelunkai_labeling_backend import models
+from spelunkai_labeling_backend.db import Database
+from spelunkai_labeling_backend.ingest import ingest_session
+
+
+def _make_frame_files(frames_dir: Path, count: int) -> None:
+    frames_dir.mkdir(parents=True, exist_ok=True)
+    for i in range(count):
+        (frames_dir / f"frame_{i:06d}.png").write_bytes(b"fake")
+
+
+def _make_database(tmp_path: Path) -> Database:
+    database = Database(f"sqlite:///{tmp_path / 'test.db'}")
+    database.init_models()
+    return database
+
+
+def test_ingest_session_inserts_all_frames(tmp_path):
+    frames_dir = tmp_path / "run01_frames"
+    _make_frame_files(frames_dir, 3)
+    database = _make_database(tmp_path)
+
+    result = ingest_session(
+        database=database,
+        frames_dir=frames_dir,
+        session_name="run01",
+        image_path_prefix="run01_frames",
+        width=1280,
+        height=720,
+    )
+
+    assert result.inserted == 3
+    assert result.already_existed == 0
+    assert result.total_seen == 3
+
+    db_session = database.session_factory()
+    try:
+        frames = db_session.query(models.Frame).order_by(models.Frame.frame_index).all()
+        assert [f.frame_index for f in frames] == [0, 1, 2]
+        assert [f.image_path for f in frames] == [
+            "run01_frames/frame_000000.png",
+            "run01_frames/frame_000001.png",
+            "run01_frames/frame_000002.png",
+        ]
+        assert frames[0].width == 1280
+        assert frames[0].height == 720
+    finally:
+        db_session.close()
+
+
+def test_ingest_session_is_idempotent_and_picks_up_new_frames(tmp_path):
+    frames_dir = tmp_path / "run01_frames"
+    _make_frame_files(frames_dir, 2)
+    database = _make_database(tmp_path)
+
+    ingest_session(database, frames_dir, "run01", "run01_frames", 1280, 720)
+
+    _make_frame_files(frames_dir, 3)  # adds frame_000002.png
+    result = ingest_session(database, frames_dir, "run01", "run01_frames", 1280, 720)
+
+    assert result.inserted == 1
+    assert result.already_existed == 2
+    assert result.total_seen == 3
+
+
+def test_ingest_session_keeps_sessions_independent(tmp_path):
+    frames_dir = tmp_path / "run01_frames"
+    _make_frame_files(frames_dir, 2)
+    database = _make_database(tmp_path)
+
+    ingest_session(database, frames_dir, "run01", "run01_frames", 1280, 720)
+    result = ingest_session(database, frames_dir, "run02", "run01_frames", 1280, 720)
+
+    assert result.inserted == 2
+    assert result.already_existed == 0
+
+
+def test_ingest_session_raises_when_no_frames_found(tmp_path):
+    frames_dir = tmp_path / "empty"
+    frames_dir.mkdir()
+    database = _make_database(tmp_path)
+
+    with pytest.raises(FileNotFoundError):
+        ingest_session(database, frames_dir, "run01", "empty", 1280, 720)