Procuria/face-sets
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Make per-clip sidecar JSONs opt-in (default off)

Browse Source 
Previously every video_target_pipeline cut wrote a <uuid>.json provenance
sidecar alongside each <uuid>.mp4. The same provenance is already in the
per-batch plan.json, so the per-clip sidecars are redundant unless a
downstream tool wants each clip self-describing in isolation.

- video_target_pipeline.py cut: new --write-sidecar flag, default off.
- run_video_pipeline.sh: new SIDECAR env var (default "no"), passes
  --write-sidecar when SIDECAR=yes.
- README + docs/analysis/video-target-preprocessing.md updated.

The 1,984 already-emitted sidecars in /mnt/x/src/vd/ct/ct_src_*/ have
been deleted (1.5 MB).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Peter
2026-04-28 12:44:27 +02:00
parent 998fa79f81
commit 7960dec350
4 changed files with 29 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
README.md
View File

@@ -343,7 +343,7 @@ clean it up over time:
| `work/consolidate_facesets.py` | Merge duplicate identities (centroid cosine sim ≥ 0.55 with confident ≥ 0.65, **complete-linkage** to defeat single-link chaining). Pulls embeddings from cache, no GPU. See `docs/analysis/identity-consolidation-and-age-extend.md`. | | `work/consolidate_facesets.py` | Merge duplicate identities (centroid cosine sim ≥ 0.55 with confident ≥ 0.65, **complete-linkage** to defeat single-link chaining). Pulls embeddings from cache, no GPU. See `docs/analysis/identity-consolidation-and-age-extend.md`. |
| `work/age_extend_001.py` | Slot newly-added PNGs into existing era buckets of `faceset_001` (anchor cosine distance ≤ 0.40 AND `|year_delta|` ≤ 5). Same anchor-fragment rule as `age_split_001.py`. | | `work/age_extend_001.py` | Slot newly-added PNGs into existing era buckets of `faceset_001` (anchor cosine distance ≤ 0.40 AND `|year_delta|` ≤ 5). Same anchor-fragment rule as `age_split_001.py`. |
| `work/dedup_optimize.py` (+ Windows `work/multiface_worker.py`) | (a) cross-family SHA256 byte-dedup, (b) within-faceset near-dup at cosine sim ≥ 0.95, (c) multi-face audit (re-detect via insightface, drop PNGs with face_count ≠ 1). Multi-face is the load-bearing roop invariant. See `docs/analysis/dedup-and-roop-optimization.md`. | | `work/dedup_optimize.py` (+ Windows `work/multiface_worker.py`) | (a) cross-family SHA256 byte-dedup, (b) within-faceset near-dup at cosine sim ≥ 0.95, (c) multi-face audit (re-detect via insightface, drop PNGs with face_count ≠ 1). Multi-face is the load-bearing roop invariant. See `docs/analysis/dedup-and-roop-optimization.md`. |
| `work/video_target_pipeline.py` (+ Windows `work/video_face_worker.py` + `work/run_video_pipeline.sh` chain) | Target-side preprocessing: scan a folder of videos → PySceneDetect shot-cuts → 2 fps frame sampling → DML face detection + embedding → IoU+embedding tracking → quality-gated segments (yaw≤75°, face≥80px, det≥0.5, ≥70% pass-rate, 1–120s duration, 2s cross-track merge gap) → ffmpeg stream-copy into UUID-named clips with sidecar JSON. Output organized into per-source subfolders. See `docs/analysis/video-target-preprocessing.md`. | | `work/video_target_pipeline.py` (+ Windows `work/video_face_worker.py` + `work/run_video_pipeline.sh` chain) | Target-side preprocessing: scan a folder of videos → PySceneDetect shot-cuts → 2 fps frame sampling → DML face detection + embedding → IoU+embedding tracking → quality-gated segments (yaw≤75°, face≥80px, det≥0.5, ≥70% pass-rate, 1–120s duration, 2s cross-track merge gap) → ffmpeg stream-copy into UUID-named clips. Output organized into per-source subfolders. Provenance sidecars are opt-in (`cut --write-sidecar` or `SIDECAR=yes` env var); the full plan is always retained in the per-batch `plan.json`. See `docs/analysis/video-target-preprocessing.md`. |
All four operate idempotently and reversibly: dropped PNGs go to All four operate idempotently and reversibly: dropped PNGs go to
`<faceset>/faces/_dropped/`, quarantined whole facesets go to `<faceset>/faces/_dropped/`, quarantined whole facesets go to

5
docs/analysis/video-target-preprocessing.md
View File

@@ -38,10 +38,11 @@ run_video_pipeline.sh (chain driver)
└─ report (HTML preview) └─ report (HTML preview)
Output: <output_dir>/<source_video_stem>/<uuid>.mp4 Output: <output_dir>/<source_video_stem>/<uuid>.mp4
/<uuid>.json (sidecar) /<uuid>.json (sidecar; opt-in via
--write-sidecar)
``` ```
`run_video_pipeline.sh` is parameterized via env vars (`WORK`, `INPUT_DIR`, `OUTPUT_DIR`, `FILTER_FROM`, `SKIP_PATTERN`, `MAX_DUR`, `IDENTITY`) so you can pin a particular batch without editing the script. `run_video_pipeline.sh` is parameterized via env vars (`WORK`, `INPUT_DIR`, `OUTPUT_DIR`, `FILTER_FROM`, `SKIP_PATTERN`, `MAX_DUR`, `IDENTITY`, `SIDECAR`) so you can pin a particular batch without editing the script. Sidecars are off by default — the per-batch `plan.json` always carries the full provenance for every clip; the `<uuid>.json` files alongside the clips are redundant and only useful if you need each clip to be self-describing in isolation.
## 3. Quality signals (matched to inswapper_128's working envelope) ## 3. Quality signals (matched to inswapper_128's working envelope)

8
work/run_video_pipeline.sh
View File

@@ -15,6 +15,7 @@
# SKIP_PATTERN regex of basenames to exclude (Python `re` syntax). Applied AFTER FILTER_FROM. # SKIP_PATTERN regex of basenames to exclude (Python `re` syntax). Applied AFTER FILTER_FROM.
# MAX_DUR score --max-dur (default 120) # MAX_DUR score --max-dur (default 120)
# IDENTITY "yes" to enable identity tagging; default "no" # IDENTITY "yes" to enable identity tagging; default "no"
# SIDECAR "yes" to emit <uuid>.json provenance sidecars; default "no"
set -e set -e
@@ -23,6 +24,7 @@ set -e
: ${OUTPUT_DIR:=/mnt/x/src/vd/ct} : ${OUTPUT_DIR:=/mnt/x/src/vd/ct}
: ${MAX_DUR:=120} : ${MAX_DUR:=120}
: ${IDENTITY:=no} : ${IDENTITY:=no}
: ${SIDECAR:=no}
mkdir -p "$WORK" "$WORK/scenes" mkdir -p "$WORK" "$WORK/scenes"
@@ -37,7 +39,7 @@ log() { echo "[$(ts)] [$PHASE] $*"; }
PHASE="setup" PHASE="setup"
log "STARTED — host=$(hostname) pid=$$ work=$WORK" log "STARTED — host=$(hostname) pid=$$ work=$WORK"
log "config: input=$INPUT_DIR output=$OUTPUT_DIR filter_from=${FILTER_FROM:-<none>} skip_pattern=${SKIP_PATTERN:-<none>} max_dur=$MAX_DUR identity=$IDENTITY" log "config: input=$INPUT_DIR output=$OUTPUT_DIR filter_from=${FILTER_FROM:-<none>} skip_pattern=${SKIP_PATTERN:-<none>} max_dur=$MAX_DUR identity=$IDENTITY sidecar=$SIDECAR"
PHASE="inventory" PHASE="inventory"
log "building subset inventory" log "building subset inventory"
@@ -110,7 +112,9 @@ log "done in $(($(date +%s)-T0))s"
PHASE="cut" PHASE="cut"
log "ffmpeg stream-copy into per-source subfolders (no --clean)" log "ffmpeg stream-copy into per-source subfolders (no --clean)"
T0=$(date +%s) T0=$(date +%s)
$PY_WSL $PIPELINE cut --plan "$WORK/plan.json" --output-dir "$OUTPUT_DIR" SIDECAR_FLAG=""
if [ "$SIDECAR" = "yes" ]; then SIDECAR_FLAG="--write-sidecar"; fi
$PY_WSL $PIPELINE cut --plan "$WORK/plan.json" --output-dir "$OUTPUT_DIR" $SIDECAR_FLAG
log "done in $(($(date +%s)-T0))s" log "done in $(($(date +%s)-T0))s"
PHASE="report" PHASE="report"

36
work/video_target_pipeline.py
View File

@@ -722,23 +722,23 @@ def cmd_cut(args):
if out_video.exists() and out_video.stat().st_size < 1024: if out_video.exists() and out_video.stat().st_size < 1024:
out_video.unlink() out_video.unlink()
continue continue
# sidecar (alongside the clip in the source-named subfolder) if args.write_sidecar:
sidecar = seg_dir / f"{seg['uuid']}.json" sidecar = seg_dir / f"{seg['uuid']}.json"
sidecar.write_text(json.dumps({ sidecar.write_text(json.dumps({
"uuid": seg["uuid"], "uuid": seg["uuid"],
"source_video": seg["video_path"], "source_video": seg["video_path"],
"source_basename": Path(seg["video_path"]).name, "source_basename": Path(seg["video_path"]).name,
"start_s": s, "end_s": seg["end_s"], "duration_s": d, "start_s": s, "end_s": seg["end_s"], "duration_s": d,
"scene_idx": seg["scene_idx"], "scene_idx": seg["scene_idx"],
"track_idxs": seg.get("track_idxs", [seg.get("track_idx")]), "track_idxs": seg.get("track_idxs", [seg.get("track_idx")]),
"member_count": seg.get("member_count"), "member_count": seg.get("member_count"),
"pass_count": seg.get("pass_count"), "pass_count": seg.get("pass_count"),
"stats": seg["stats"], "stats": seg["stats"],
"identity_tag": seg["identity_tag"], "identity_tag": seg["identity_tag"],
"identity_sim": seg["identity_sim"], "identity_sim": seg["identity_sim"],
"thresholds": plan["thresholds"], "thresholds": plan["thresholds"],
}, indent=2)) }, indent=2))
sidecars.append(sidecar) sidecars.append(sidecar)
n_done += 1 n_done += 1
print(f"[cut] {n_done} clips written, {n_err} errors -> {out_dir}", file=sys.stderr) print(f"[cut] {n_done} clips written, {n_err} errors -> {out_dir}", file=sys.stderr)
@@ -901,6 +901,8 @@ def main():
cu.add_argument("--force", action="store_true") cu.add_argument("--force", action="store_true")
cu.add_argument("--clean", action="store_true", cu.add_argument("--clean", action="store_true",
help="remove prior UUID-named clips before cutting (preserves non-UUID files)") help="remove prior UUID-named clips before cutting (preserves non-UUID files)")
cu.add_argument("--write-sidecar", action="store_true",
help="emit <uuid>.json provenance sidecar alongside each clip (default off)")
cu.set_defaults(func=cmd_cut) cu.set_defaults(func=cmd_cut)
rp = sub.add_parser("report") rp = sub.add_parser("report")