Peach: Local-first video metadata scraper and media manager.
"""逐张核对封面尺寸、来源、版次、人脸与正封框,保存可重放的核查清单。"""
from __future__ import annotations
import argparse
import csv
import json
import shutil
import sys
from collections import Counter
from pathlib import Path
sys.path.insert(0, str(Path(__file__).resolve().parents[1] / 'src'))
from PIL import Image
from peach import jav_poster_crop as crop
from peach.cover_artwork import FACE_ALGORITHM_VERSION, metadata_snapshots, face_record as detect_face
from peach.community_catalog import origin_of
from peach.fsutil import atomic_write_text
from peach.library_processing import _official_evidence
def read(path):
try:
data = json.loads(path.read_text(encoding='utf-8'))
return data if isinstance(data, dict) else {}
except (OSError, ValueError):
return {}
def snapshots(root, code):
return metadata_snapshots(root, code)
def current_face(path, size, title, detector):
"""按算法版本与源图尺寸核对缓存,需要时从整图重新检测。"""
record = read(path.with_suffix('.face.json'))
if detector is not None and (record.get('version') != FACE_ALGORITHM_VERSION or record.get('px') != list(size)):
from peach.face_detect import decode
return detect_face(decode(path.read_bytes()), detector, code=path.stem, title=title) or {}
return record
def inspect(path, metadata, *, detector=None):
from peach.library_processing import artwork_title
with Image.open(path) as image:
width, height = image.size
image.verify()
saved = crop.read_sidecar(path) or {}
evidence = read(path.with_suffix('.scraping.json'))
known = snapshots(metadata, path.stem)
previous_title = str(saved.get('title') or evidence.get('title') or '')
title = artwork_title([*known, ('installed', {'title': previous_title, 'genres': [previous_title]})])
if crop.is_vr_title(previous_title):
title = previous_title
face_record = current_face(path, (width, height), title, detector)
face = face_record.get('face') if face_record.get('px') == [width, height] else None
full = crop.face_crosses_spine(width, height, face)
layout = 'still' if crop.is_scene_cover(path.stem, title) or face_record.get('layout') == 'still' else ''
proposed = saved if crop.is_manual(saved) and crop.is_current(saved, width, height) else crop.crop_record(
path.stem, width, height, crop.file_gradient(path), title=title, face=face, layout=layout)
if full and proposed['box']['method'] == crop.NONE:
layout = 'still'
if layout == 'still' and not crop.is_manual(proposed):
proposed['box'] = {'x0': 0, 'y0': 0, 'x1': width, 'y1': height, 'method': crop.NONE}
origin = origin_of(str(evidence.get('source_url') or ''))
reasons = []
if origin in ('javdb', 'javbus'):
reasons.append('community_source')
if not evidence.get('source_url'):
reasons.append('missing_provenance')
if max(width, height) < 1600:
reasons.append('low_resolution')
if not crop.is_current(saved, width, height):
reasons.append('stale_frame')
if face_record.get('version') != FACE_ALGORITHM_VERSION:
reasons.append('stale_face')
if saved.get('box') != proposed.get('box'):
reasons.append('crop_changed')
if proposed['box']['method'] == crop.RATIO:
reasons.append('ratio_fallback')
if layout != str(face_record.get('layout') or '') and not crop.is_manual(saved):
reasons.append('layout_changed')
official = _official_evidence(known)
row = dict(code=path.stem, width=width, height=height, origin=origin,
method=proposed['box']['method'], old_x0=(saved.get('box') or {}).get('x0'),
x0=proposed['box']['x0'], layout=layout, official_candidates=len(official.candidates) if official else 0,
reasons=';'.join(reasons))
return row, proposed, face_record, layout
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--data-root', type=Path, required=True)
parser.add_argument('--report', type=Path, required=True)
parser.add_argument('--apply', action='store_true', help='备份并更新可重建的取景边车')
parser.add_argument('--refresh-faces', action='store_true', help='按当前人脸算法重算过期缓存')
args = parser.parse_args()
args.report.mkdir(parents=True, exist_ok=True)
rows, errors = [], []
covers = sorted((args.data_root / 'generated/covers').glob('*.jpg'))
detector = None
if args.refresh_faces:
from peach.face_detect import FaceDetector
detector = FaceDetector(allow_download=False)
for index, path in enumerate(covers, 1):
try:
row, poster, face, layout = inspect(path, args.data_root / 'sources/library-metadata', detector=detector)
rows.append(row)
if args.apply:
changes = {path.with_suffix('.poster.json'): poster}
if face and not crop.is_manual(poster):
face = {**face}
face.pop('layout', None)
if layout:
face['layout'] = layout
changes[path.with_suffix('.face.json')] = face
for target, data in changes.items():
if read(target) == data:
continue
backup = args.report / 'sidecar-backup' / target.name
if target.exists() and not backup.exists():
backup.parent.mkdir(parents=True, exist_ok=True)
shutil.copy2(target, backup)
atomic_write_text(target, json.dumps(data, ensure_ascii=False))
except (OSError, ValueError, KeyError) as error:
errors.append({'code': path.stem, 'error': str(error)})
if index % 100 == 0:
print(f'{index}/{len(covers)}', flush=True)
if rows:
with (args.report / 'covers.csv').open('w', newline='', encoding='utf-8-sig') as output:
writer = csv.DictWriter(output, fieldnames=list(rows[0]))
writer.writeheader()
writer.writerows(rows)
summary = dict(total=len(covers), decoded=len(rows), errors=errors,
origins=dict(Counter(row['origin'] for row in rows)),
reasons=dict(Counter(reason for row in rows for reason in row['reasons'].split(';') if reason)),
applied=args.apply)
atomic_write_text(args.report / 'summary.json', json.dumps(summary, ensure_ascii=False, indent=2))
print(json.dumps(summary, ensure_ascii=False), flush=True)
return 1 if errors else 0
if __name__ == '__main__':
raise SystemExit(main())