def dumpWithinContext(dirs):
dumping_functions = [('all_segments.tab', dumpSegments),
('all_canonicals.tab', dumpCanonicals)]
for dumping_function in dumping_functions:
log(INFO, "Storing %s" % (dumping_function[0]))
filename = os.path.join(dirs[0], dumping_function[0])
dumping_function[1](filename)
def extract_all(self, job_path='', results_dir ='.'):
"""
Extract all videos found in dir and below.
:param job_path: path of a
:param results_dir: store the results in the given directory
:return: True if success, if not, the output_dir content should NOT be changed (clean up)
"""
import shutil
job_path = os.path.abspath(job_path)
data = self._compose_job_definition(job_path)
str_ = json.dumps(data, indent=4, sort_keys=False, separators=(',', ': '), ensure_ascii=False)
with open(os.path.join(results_dir, 'job.json'), 'w') as new_file:
new_file.write(str_)
shutil.copy(os.path.join(str(self.extractor_location),'mime.types'),'.')
with open(os.path.join(results_dir, 'cineast.json'), 'w') as new_file:
with open(os.path.join(self.extractor_location, 'cineast.json'), 'r') as fp:
data = json.load(fp)
data["extractor"]["outputLocation"] = results_dir
str_ = json.dumps(data, indent=4, sort_keys=False, separators=(',', ': '), ensure_ascii=False)
new_file.write(str_)
command = ['java', '-Xmx6G', '-Xms6G', '-jar', os.path.join(str(self.extractor_location),'cineast.jar'),
'--job', os.path.abspath(os.path.join(results_dir, 'job.json')), '--config', os.path.abspath(os.path.join(results_dir, 'cineast.json'))]
p = Popen(command, stderr=PIPE, stdout=PIPE)
stdout, stderr = p.communicate()
log(INFO,stdout)
log(ERROR,stderr)
if p.returncode == 0:
try:
os.remove(os.path.abspath(os.path.join(results_dir, 'job.json')))
os.remove( os.path.abspath(os.path.join(results_dir, 'cineast.json')))
except:
pass
return p.returncode == 0
def unwrap(path, location=None, check_md5=True):
"""
:param path:
:param location:
:param check_md5:
:return: None if error
"""
if location is None:
location = os.path.split(os.path.abspath(path))[0]
newname = os.path.join(location, os.path.splitext(os.path.basename(path))[0])
if os.path.exists(newname):
return newname
if not os.path.exists(location):
mkdirs(location)
with open(os.path.normpath(path), 'rb') as fp:
scanner = Scanner(os.path.normpath(path), fp)
if scanner.readHeader():
newname = scanner.writeContents(location)
if check_md5:
md5 = scanner.md5(newname)
if scanner.data['data_md5'] != md5:
log(ERROR,'corrupt {}'.format(path))
os.remove(newname)
return None
return newname
def loadObjects(object_type, filename):
from aida.media.object import MediaObject, MediaObjectIdentifier
loader = MediaObjectIdentifier()
data = loader.load(filename, object_type)
matches = {}
preloads = {}
premappings = set()
for x in db.session.query(Segment.frame_file_name, Segment.id).all():
matches[x.frame_file_name] = x.id
for x in db.session.query(SegmentObject.segment_id,
SegmentObject.bounding_box,
SegmentObject.id).all():
preloads[x.segment_id + SegmentObject.bounding_box] = x.id
for x in db.session.query(SegmentObjectMatch.to_obj,
SegmentObjectMatch.from_obj).all():
premappings.add(x.to_obj + x.from_obj)
mappings = {}
for v in data.values():
item = None
levels = 3
while levels > 0 and item is None:
search_id = v.id.split(os.path.sep)[-levels:]
key = os.path.sep.join(search_id)
key = os.path.normpath(key)
if key in matches:
item = matches[key]
levels -= 1
if item is None:
log(ERROR, "Cannot find file {} ".format(v.id))
elif item + v.text_bounding_box() in preloads:
mappings[v.id] = preloads[item + v.text_bounding_box()]
continue
else:
log(INFO, "File {} ".format(v.id))
new_id = uuid.uuid4().urn
db.session.add(
SegmentObject(id=new_id,
bounding_box=v.text_bounding_box(),
object_type=v.object_type,
segment_id=item))
mappings[v.id] = new_id
db.session.commit()
commit_time = 1000
for v in data.values():
for rel in v.relationships:
new_id = uuid.uuid4().urn
#print("%s %s" % (mappings[rel[0]], mappings[v.id]))
if (mappings[rel[0]] + mappings[v.id]) in premappings:
continue
db.session.add(
SegmentObjectMatch(id=new_id,
to_obj=mappings[rel[0]],
from_obj=mappings[v.id],
measure=rel[1]))
commit_time -= 1
if commit_time == 0:
commit_time = 1000
db.session.commit()
db.session.commit()
def _load_provider(self, app, provider):
log(INFO, "Loading %s" % (provider.__class__.__name__))
for k, v in provider.config(
external_config=self.external_config).items():
app.config[k] = v
for dependency in provider.dependencies():
if self.get_provider_by_name(dependency.__name__) is None:
self.providers.append(dependency())
self._load_provider(app, self.providers[-1])
def unZip(dir, item, destination):
with ZipFile(os.path.join(dir, item), 'r') as inzip:
for info in inzip.infolist():
if info.filename[-1] == '/':
continue
if info.filename.endswith('ldcc'):
info.filename = os.path.basename(info.filename)
log(INFO,'extracting: ' + str(info.filename))
inzip.extract(info, destination)
yield os.path.join(destination,info.filename)
def findFFMPEG():
try:
p = Popen('ffmpeg', stderr=PIPE)
p.communicate()
return 'ffmpeg'
except FileNotFoundError as e:
try:
return os.environ['FFMPEG_BIN']
except KeyError:
log(
ERROR,
'Cannot find ffmpeg. Environment variable FFMPEG_BIN not set')
def extract(self, dirs, destination):
extractor = KeyFrameExtractor(self.cineast)
for dir in dirs:
ldc.unZipAll(dir,
filters=['mp4'],
location=destination,
unwrap_files=True,
cleanup=True)
object_json = os.path.join(destination, 'json',
'cineast_multimediaobject.json')
segment_json = os.path.join(destination, 'json',
'cineast_segment.json')
representative_json = os.path.join(
destination, 'json', 'cineast_representativeframes.json')
if not os.path.exists(object_json) or not os.path.exists(segment_json):
extractor.extract_all(job_path=destination,
results_dir=destination)
if not os.path.exists(object_json) or not os.path.exists(
segment_json) or not os.path.exists(representative_json):
log(ERROR, 'key frame extraction failed')
def default(self, obj):
def _prim_check(obj):
if isinstance(obj, (datetime, date)):
return obj.isoformat()
elif isinstance(obj.__class__,
InstrumentedList) or type(obj) == InstrumentedList:
return [self.default(item) for item in obj]
else:
return obj
def _complex_check(obj):
if isinstance(obj.__class__, DeclarativeMeta):
# an SQLAlchemy class
fields = {}
valid_values = obj.__json__() if hasattr(obj,
'__json__') else None
for field in [x for x in dir(obj) if not x.startswith('_') and x != 'metadata' and \
(valid_values is None or x in valid_values)]:
data = obj.__getattribute__(field)
if isinstance(data, BaseQuery):
continue
result = _complex_check(data)
fields[field] = result
return fields
if isinstance(obj.__class__,
InstrumentedList) or type(obj) == InstrumentedList:
return [_complex_check(item) for item in obj]
return _prim_check(obj)
try:
return _complex_check(obj)
except TypeError as e:
log(ERROR, str(e))
return None
def getReferencedFramesGivenSegments(filename,
segments=[],
chunk_size=20,
alternate_directory=None):
"""
:param filename:
:param segments:
:param chunk_size:
:return:
@rtype: list (VideoSegment)
"""
chunks = [
segments[i:i + chunk_size] for i in range(0, len(segments), chunk_size)
]
frames_done = 0
for chunk in chunks:
expression = [
'eq(n\,%d)' % (segment.representative_frame_number)
for segment in chunk
]
base = os.path.splitext(filename)[0]
if alternate_directory is not None:
base = os.path.join(alternate_directory, os.path.basename(base))
expression_str = '+'.join(expression)
if ffmpeg_command is None or len(ffmpeg_command) == 0:
log(ERROR, 'FFMPEG COMMAND NOT FOUND')
command = [
ffmpeg_command, '-i', filename, '-vf', 'select=' + expression_str,
'-vsync', '0', '-start_number',
str(frames_done), '{}_%06d.png'.format(base)
]
log(INFO, ' '.join(command))
p = Popen(command, stderr=PIPE)
stderr = p.communicate()
if p.returncode == 0:
for i in range(len(chunk)):
segment = segments[frames_done]
segment_file = '%s%06d.png' % (base + '_', frames_done)
segment.set_sample_img(cv2.imread(segment_file), segment_file)
yield segment
else:
log(ERROR, stderr)
def loadWithinContext(dirs):
import re
from functools import partial
# map to loading functions
# order by the order loading.
# None is indicative of future use!
loading_function_mappings = [
("hypothesis_info\.tab", loadHypothesis),
("media_list.tab", loadMedia),
("parent_children\.tab", loadParentChildDetails),
("twitter_info\.tab", None),
("uid_info\.tab", None),
("uids_missing_video\.tab", None),
("T.*KB\.tab", loadKB),
("T.*ent_mentions\.tab", loadKBEntities), # yes, same file
("T.*ent_mentions\.tab", loadMediaEntities), # yes, same file
("T.*ent_mentions\.tab", loadEntities), # yes, same file
("T.*ent_mentions\.tab", loadEntityMentions),
("T.*evt_mentions\.tab", None),
("T.*evt_slots\.tab", None),
("T.*rel_mentions\.tab", None),
("T.*rel_slots\.tab", None),
("T.*hypothesis\.tab", None),
("rel_evt_slot_mapping_table\.tab", None),
("doc_lang\.tab", None),
(".*canonicals\.tab", loadCanonicals),
(".*segments\.tab", loadSegments),
(".*_multimediaobject\.json", loadKeyFrames),
("image.json", loadImages),
("ocr.txt", partial(loadOCR, 'inf')),
("object_faces.*\.json", partial(loadObjects, 'fac'))
]
loading_function_mappings = [(re.compile(mapping[0]), mapping[1])
for mapping in loading_function_mappings]
# find the files; organize by name
files_by_match = [[] for pos in range(len(loading_function_mappings))]
count = 0
for dir in dirs:
log(INFO, "Scanning directory %s" % (dir))
for root, walked_dirs, files in os.walk(dir):
for name in files:
for pos in range(len(loading_function_mappings)):
if loading_function_mappings[pos][1] is None:
continue
if loading_function_mappings[pos][0].match(
name) is not None:
count += 1
files_by_match[pos].append(os.path.join(root, name))
log(INFO, "Loading %d files" % (count))
for pos in range(len(loading_function_mappings)):
mapping = loading_function_mappings[pos]
matched_files = files_by_match[pos]
# have a loading function?
if mapping[1] is not None:
# load each matching the mapping's name
for matched_file in matched_files:
log(INFO, "Loading %s" % (matched_file))
try:
mapping[1](matched_file)
except Exception as e:
log(ERROR,
"Failed to load %s: %s" % (matched_file, str(e)))
exc_type, exc_value, exc_traceback = exc_info()
print_tb(exc_traceback, limit=1, file=stdout)
raise e
db.session.commit()
log(INFO, "Loading complete")
def __init__(self, languages=['eng']):
self.languages = languages
log(INFO, 'configured OCR with languages %s' % ', '.join(languages))