File size: 2,824 Bytes
94033a8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 | """Offline task index reader. No model calls and no execution of image instructions."""
import csv
import json
from collections import Counter
from pathlib import Path
from urllib.parse import urlsplit
EXPECTED = {'GUI': 209, 'Document': 200, 'Web': 200, 'Mobile': 283, 'Video': 191}
def local_file(root, value):
root = Path(root).resolve()
if not value or urlsplit(value).scheme or Path(value).is_absolute():
raise ValueError('Expected a root-relative local file')
p = (root / value).resolve()
if not p.is_relative_to(root) or not p.is_file():
raise ValueError('Missing or out-of-root file')
return p
def load_tasks(root=None, domain=None):
"""Yield manifest dictionaries with parsed history lists, preserving root-relative paths."""
root = Path(root or Path(__file__).parent).resolve()
with (root / 'task_manifest.csv').open(newline='', encoding='utf-8') as f:
rows = list(csv.DictReader(f))
if Counter(r['domain'] for r in rows) != Counter(EXPECTED):
raise ValueError('Unexpected benchmark membership')
if len({r['record_id'] for r in rows}) != len(rows):
raise ValueError('Duplicate record IDs')
if domain is not None and domain not in EXPECTED:
raise ValueError('Unknown domain')
for row in rows:
if domain is not None and row['domain'] != domain:
continue
row['history_media'] = json.loads(row.pop('history_media_json'))
for key in ('annotation_file', 'input_media', 'clean_media'):
if row[key]:
local_file(root, row[key])
for h in row['history_media']:
local_file(root, h)
yield row
def load_annotation(root, row):
"""Return the original task record; Video returns the video ID, QA and video info."""
p = local_file(root, row['annotation_file'])
if row['annotation_line'] != '':
j = json.loads(p.read_text().splitlines()[int(row['annotation_line']) - 1])
q = j['qa'][int(row['qa_index'])]
if str(j['key']) + ':' + str(q['uid']) != row['task_id']:
raise ValueError('Video locator no longer matches task')
return {'video_id': j['key'], 'qa': q, 'video_info': j['video_info']}
j = json.loads(p.read_text())
if row['annotation_index'] != '':
j = j[int(row['annotation_index'])]
if str(j['task_id']) != row['task_id']:
raise ValueError('Annotation locator no longer matches task')
return j
if __name__ == '__main__':
root = Path(__file__).parent
rows = list(load_tasks(root))
for row in rows:
load_annotation(root, row)
print(json.dumps({'tasks': len(rows), 'domains': dict(Counter(r['domain'] for r in rows)),
'all_indexed_files_exist': True, 'all_annotation_locators_match': True}, indent=2))
|