def test_create_an_entity_with_image_justification_and_vector(self):
g = aifutils.make_graph()
g.bind('ldcOnt', SEEDLING_TYPES_NIST.uri)
# every AIF needs an object for the system responsible for creating it
system = aifutils.make_system_with_uri(
g, "http://www.test.edu/testSystem")
# it doesn't matter what URI we give entities, events, etc. so long as they are
# unique
entity = aifutils.make_entity(g, "http://www.test.edu/entities/1",
system)
# in order to allow uncertainty about the type of an entity, we don't mark an
# entity's type directly on the entity, but rather make a separate assertion for it
# its URI doesn't matter either
type_assertion = aifutils.mark_type(
g, "http://www.test.org/assertions/1", entity,
SEEDLING_TYPES_NIST.Person, system, 1.0)
# the justification provides the evidence for our claim about the entity's type
# we attach this justification to both the type assertion and the entity object
# itself, since it provides evidence both for the entity's existence and its type.
# in TA1 -> TA2 communications, we attach confidences at the level of justifications
# let's suppose we also have evidence from an image
bb1 = Bounding_Box((123, 45), (167, 98))
aifutils.mark_image_justification(g, [entity, type_assertion],
"NYT_ENG_20181231_03", bb1, system,
0.123)
# also we can link this entity to something in an external KB
aifutils.link_to_external_kb(g, entity, "freebase.FOO", system, .398)
vec = {
"vector_type": "http://www.test.edu/systemX/personVector",
"vector_data": [2.0, 7.5, 0.2, 8.1]
}
# let's mark our entity with some arbitrary system-private data. You can attach such data
# to nearly anything
aifutils.mark_private_data_with_vector(g, entity, system, vec)
self.new_file(
g, "test_create_an_entity_with_image_justification_and_vector.ttl")
self.dump_graph(
g, "Example of entity with image justification and vector")
def transferAIF(p_id):
#for k,p_id in enumerate(parent_dict):
# Todo test
# if (k<8):
# continue
# print('k',k,p_id)
g = Graph()
#load rpi graph if exists
if p_id in txt_mention_ttl_list:
turtle_path = os.path.join(txt_mention_ttl_path, p_id + '.ttl')
turtle_content = open(turtle_path).read()
g.parse(data=turtle_content, format='n3')
#load and merge cu graph if exists
if p_id in cu_ttl_list:
turtle_path = os.path.join(cu_ttl_path, p_id + '.ttl')
turtle_content = open(turtle_path).read()
g.parse(data=turtle_content, format='n3')
#load and merge cu graph for instance matching if exists
if p_id in cu_ttl_ins_list:
turtle_path = os.path.join(cu_ttl_ins_path, p_id + '.ttl')
turtle_content = open(turtle_path).read()
g.parse(data=turtle_content, format='n3')
sys_instance_matching = aifutils.make_system_with_uri(
g, cu_pref + 'Systems/Instance-Matching/ResNet152')
sys_grounding = aifutils.make_system_with_uri(
g, cu_pref + 'Systems/Grounding/ELMo-PNASNET')
usc_sys_grounding = aifutils.make_system_with_uri(
g, usc_pref + 'Systems/ZSGrounder')
#find vision and text entities
sbj_all = set(g.subjects())
img_entities = {}
keyframe_entities = {}
ltf_entities = {}
for sbj in sbj_all:
sbj_name = sbj.toPython()
if cu_objdet_pref in sbj_name:
if sbj.__class__ == rdflib.term.URIRef:
if 'JPG' in sbj_name:
img_id = '/'.join(sbj_name.split('/')[-2:])
img_entities[img_id] = sbj
elif 'Keyframe' in sbj_name:
kfrm_id = '/'.join(sbj_name.split('/')[-2:])
keyframe_entities[kfrm_id] = sbj
elif rpi_entity_pref in sbj_name:
if sbj.__class__ == rdflib.term.URIRef and rpi_entity_pref in sbj_name:
ltf_entities[sbj_name] = sbj
# Done
# if p_id in []:#['IC0011TIB']:
# continue
# print('k',k,p_id)
# if (g==None):
# print('p_id', k, p_id)
##adding private data to entities for cu grounding
#images
for key in img_entities:
dtype = 'jpg'
#instance features
ftype = 'instance'
data_instance = get_features(key, dtype, ftype)
#semantic features
ftype = 'semantic'
data_semantic = get_features(key, dtype, ftype)
#aggregation
j_d_i = json.dumps({'columbia_vector_instance_v1.0': data_instance})
j_d_s = json.dumps({'columbia_vector_grounding_v1.0': data_semantic})
entity = img_entities[key]
aifutils.mark_private_data(g, entity, j_d_i, sys_instance_matching)
aifutils.mark_private_data(g, entity, j_d_s, sys_grounding)
#keyframes
for key in keyframe_entities:
dtype = 'keyframe'
#instance features
ftype = 'instance'
data_instance = get_features(key, dtype, ftype)
#semantic features
ftype = 'semantic'
data_semantic = get_features(key, dtype, ftype)
#aggregation
j_d_i = json.dumps({'columbia_vector_instance_v1.0': data_instance})
j_d_s = json.dumps({'columbia_vector_grounding_v1.0': data_semantic})
entity = keyframe_entities[key]
aifutils.mark_private_data(g, entity, j_d_i, sys_instance_matching)
aifutils.mark_private_data(g, entity, j_d_s, sys_grounding)
cnt_img = {}
cnt_boxO = {}
cnt_ltf = {}
#add text features, grounding, linking
for key in ltf_entities:
if key not in grounding_dict:
continue
entity_name = None
USC_GROUNDING = 'usc_vision' in grounding_dict[key][
'grounding'].values()
if not USC_GROUNDING:
# print('our grounding')
#text features
j_d_t = json.dumps({
'columbia_vector_text_v1.0':
grounding_dict[key]['textual_features'].tolist()
})
entity_ltf = ltf_entities[key]
aifutils.mark_private_data(g, entity_ltf, j_d_t, sys_grounding)
#type and name of entity to be linked
type_rdf = grounding_dict[key]['type_rdf']
entity_name = grounding_dict[key]['name']
grndg_file_type = grounding_dict[key]['source_type']
if entity_name is None:
continue
#keep track of entities with same names for avoiding clustering overlap
if entity_name in cnt_ltf:
cnt_ltf[entity_name] += 1
else:
cnt_ltf[entity_name] = 1
clstr_prot_flag = False #cluster obj for entity_ltf not created yet
#adding grounding bboxes as new entities
for img_id in grounding_dict[key]['grounding']:
if img_id == 'system':
continue
grnd = grounding_dict[key]['grounding'][img_id]
for ii, bbox in enumerate(grnd['bbox']):
if img_id in cnt_img: #to keep track of cnt of bbox of same image
cnt_img[img_id] += 1
else:
cnt_img[img_id] = 1
#add grounding bbox as entity
score = grnd['bbox_score'][ii]
if not USC_GROUNDING:
type_eid = cu_grndg_type_pref + f"{grndg_file_type}/{img_id.split('.')[0]}/{cnt_img[img_id]}/ERE"
ent_eid = cu_grndg_ent_pref + f"{grndg_file_type}/{img_id.split('.')[0]}/{cnt_img[img_id]}"
entity_grnd = aifutils.make_entity(g, ent_eid,
sys_grounding)
type_assertion = aifutils.mark_type(
g, type_eid, entity_grnd, type_rdf, sys_grounding,
score)
elif USC_GROUNDING:
type_eid = usc_grndg_type_pref + f"{grndg_file_type}/{img_id.split('.')[0]}/{cnt_img[img_id]}/ERE"
ent_eid = usc_grndg_ent_pref + f"{grndg_file_type}/{img_id.split('.')[0]}/{cnt_img[img_id]}"
entity_grnd = aifutils.make_entity(g, ent_eid,
usc_sys_grounding)
type_assertion = aifutils.mark_type(
g, type_eid, entity_grnd, type_rdf, usc_sys_grounding,
score)
# Done:
# 1. add if for the branches for image and keyframe.
# 2. add aifutils.mark_keyframe_video_justification
# 3. check output
# aifutils.mark_keyframe_video_justification(g, [entity, type_assertion], "NYT_ENG_20181231_03", "keyframe ID",
# bb2, system, 0.234)
# source: HC0005BR6_23
# print(img_id)
# Done:
# merge usc_grounding dict
# add usc_grounding entities and clusters
# Test
# print("type_assertion",type_assertion, img_id)
bb = Bounding_Box((bbox[0], bbox[1]), (bbox[2], bbox[3]))
if not USC_GROUNDING:
if 'JPG' in type_assertion:
imgid = img_id.split('.')[0]
justif = aifutils.mark_image_justification(
g, [entity_grnd, type_assertion], imgid, bb,
sys_grounding, score)
#
elif 'Keyframe' in type_assertion:
imgid = img_id.split('.')[0].split('_')[0]
kfid = img_id.split('.')[0].split(
'_'
)[1] # it should be keyframe image id or keyframe number
justif = aifutils.mark_keyframe_video_justification(g, [entity, type_assertion], imgid, kfid, \
bb, sys_grounding, score)
elif USC_GROUNDING:
imgid = img_id.split('.')[0]
justif = aifutils.mark_image_justification(
g, [entity_grnd, type_assertion], imgid, bb,
usc_sys_grounding, score)
else:
print(
'[Merge Error] in Main Body: the type_assertion is wrong'
)
aifutils.add_source_document_to_justification(g, justif, p_id)
aifutils.mark_informative_justification(g, entity_grnd, justif)
if not USC_GROUNDING:
grounding_features = grnd['grounding_features'][ii].tolist(
)
instance_features = grnd['instance_features'][ii].tolist()
#add private data to this very bbox entity
j_d_g = json.dumps(
{'columbia_vector_grounding_v1.0': grounding_features})
j_d_i = json.dumps(
{'columbia_vector_instance_v1.0': instance_features})
aifutils.mark_private_data(g, entity_grnd, j_d_g,
sys_grounding)
aifutils.mark_private_data(g, entity_grnd, j_d_i,
sys_instance_matching)
#### add clusters
# Grounding Cluster
# Done: filtering about punctuation
# translator = str.maketrans(string.punctuation, '_'*len(string.punctuation),'' )
# 'entity_name'.translate(translator)
entity_name_tmp = '_'.join(entity_name.split(' '))
entity_name_in_IRI = "".join(
x if x.isalpha() or x.isdigit() or x == '_' else '-'
for x in entity_name_tmp)
# '_'.join(entity_name.split(' '))
#gbbox entity to rpi entity
if not USC_GROUNDING:
if not clstr_prot_flag: #create cluster if not present
clst_eid = cu_grndg_clstr_txt_pref + f"{entity_name_in_IRI}/{cnt_ltf[entity_name]}"
clusterObj = aifutils.make_cluster_with_prototype(
g, clst_eid, entity_ltf, sys_grounding)
clstr_prot_flag = True
#cluster current bbox with current ltf_entity
score = grnd['men-img-score'][ii]
aifutils.mark_as_possible_cluster_member(
g, entity_grnd, clusterObj, score, sys_grounding)
# Done: add prototype as member
aifutils.mark_as_possible_cluster_member(
g, entity_ltf, clusterObj, 1, sys_grounding)
elif USC_GROUNDING:
if not clstr_prot_flag: #create cluster if not present
clst_eid = usc_grndg_clstr_txt_pref + f"{entity_name_in_IRI}/{cnt_ltf[entity_name]}"
clusterObj = aifutils.make_cluster_with_prototype(
g, clst_eid, entity_ltf, usc_sys_grounding)
clstr_prot_flag = True
#cluster current bbox with current ltf_entity
score = grnd['men-img-score'][ii]
aifutils.mark_as_possible_cluster_member(
g, entity_grnd, clusterObj, score, usc_sys_grounding)
# Done: add prototype as member
aifutils.mark_as_possible_cluster_member(
g, entity_ltf, clusterObj, 1, usc_sys_grounding)
# BoundingBox Overlap Cluster (Instance Matching)
#gbbox entity to objdet entity for instance matching
if not USC_GROUNDING:
clstr_prot_b2b_flag = False
for jj, img_id_link in enumerate(
grnd['link_ids'][ii]): #for all objdet bboxes
if img_id_link in img_entities:
entity_link_img = img_entities[img_id_link]
elif img_id_link in keyframe_entities:
entity_link_img = keyframe_entities[img_id_link]
else:
continue
if img_id in cnt_boxO: #to keep track of cnt of bbox overlap for same image
cnt_boxO[img_id] += 1
else:
cnt_boxO[img_id] = 1
if not clstr_prot_b2b_flag:
clst_b2b_eid = cu_grndg_clstr_img_pref + f"{img_id.split('.')[0]}/{cnt_boxO[img_id]}"
clusterObj_b2b = aifutils.make_cluster_with_prototype(
g, clst_b2b_eid, entity_grnd,
sys_grounding) # sys_instance_matching
clstr_prot_b2b_flag = True
score = grnd['link_scores'][ii][
jj] #IoU of grnd bbox and objdet bbox
aifutils.mark_as_possible_cluster_member(
g, entity_link_img, clusterObj_b2b, score,
sys_grounding) # sys_instance_matching
# Done: add prototype as member
aifutils.mark_as_possible_cluster_member(
g, entity_grnd, clusterObj_b2b, 1,
sys_grounding) # sys_instance_matching
# Check Point: merged_ttl_D2
# /data/bobby/AIDA/M18_copy/data/merged_ttl/merged_ttl_D2/
# IC0011VEA.ttl
# GroundingBox
with open(os.path.join(merged_graph_path, p_id + '.ttl'), 'w') as fout:
serialization = BytesIO()
g.serialize(destination=serialization, format='turtle')
fout.write(serialization.getvalue().decode('utf-8'))
#sys.stdout.write('Key {}/{} \r'.format(k,len(parent_dict)))
sys.stdout.flush()
def test_create_compound_justification(self):
g = aifutils.make_graph()
system = aifutils.make_system_with_uri(g, "http://www.test.edu/system")
event = aifutils.make_event(
g,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#V779961.00010",
system)
event_type_assertion = aifutils.mark_type(
g,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#assertion-1",
event, SEEDLING_TYPES_NIST['Personnel.Elect'], system, 1.0)
# create the two entities involved in the event
putin = aifutils.make_entity(
g,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#E781167.00398",
system)
person_type_assertion = aifutils.mark_type(
g,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#assertion-2",
putin, SEEDLING_TYPES_NIST.Person, system, 1.0)
russia = aifutils.make_entity(
g,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#E779954.00004",
system)
gpe_type_assertion = aifutils.mark_type(
g,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#assertion-3",
russia, SEEDLING_TYPES_NIST.GeopoliticalEntity, system, 1.0)
# link those entities to the event
electee_argument = aifutils.mark_as_argument(
g, event, SEEDLING_TYPES_NIST['Personnel.Elect_Elect'], putin,
system, 0.785,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#assertion-4"
)
place_argument = aifutils.mark_as_argument(
g, event, SEEDLING_TYPES_NIST['Personnel.Elect_Place'], russia,
system, 0.589,
"https://tac.nist.gov/tracks/SM-KBP/2019/LdcAnnotations#assertion-5"
)
# the justification provides the evidence for our claim about the entity's type
# we attach this justification to both the type assertion and the entity object itself, since it provides
# evidence both for the entity's existence and its type.
# in TA1 -> TA2 communications, we attach confidences at the level of justifications
text_justification = aifutils.make_text_justification(
g, "NYT_ENG_20181231", 42, 143, system, 0.973)
aifutils.mark_justification(g, person_type_assertion,
text_justification)
aifutils.mark_justification(g, putin, text_justification)
aifutils.add_source_document_to_justification(
g, text_justification, "NYT_PARENT_ENG_20181231_03")
bb1 = Bounding_Box((123, 45), (167, 98))
# let's suppose we also have evidence from an image
image_justification = aifutils.make_image_justification(
g, "NYT_ENG_20181231_03", bb1, system, 0.123)
bb2 = Bounding_Box((234, 56), (345, 101))
# and also a video where the entity appears in a keyframe
keyframe_video_justification = aifutils.make_keyframe_video_justification(
g, "NYT_ENG_20181231_03", "keyframe ID", bb2, system, .0234)
#and also a video where the entity does not appear in a keyframe
shot_video_justification = aifutils.make_shot_video_justification(
g, "SOME_VIDEO", "some shot ID", system, 0.487)
# and even audio!
audio_justification = aifutils.make_audio_justification(
g, "NYT_ENG_201181231", 4.566, 9.876, system, 0.789)
# combine all justifications into single justifiedBy triple with new confidence
aifutils.mark_compound_justification(g, [electee_argument], [
text_justification, image_justification,
keyframe_video_justification, shot_video_justification,
audio_justification
], system, .321)
aifutils.mark_compound_justification(
g, [place_argument], [text_justification, image_justification],
system, 0.543)
self.new_file(g, "test_create_compound_justification.ttl")
self.dump_graph(g, "Example of compound justification")
def test_create_an_entity_with_all_justification_types_and_confidence(
self):
g = aifutils.make_graph()
g.bind('ldcOnt', SEEDLING_TYPES_NIST.uri)
# every AIF needs an object for the system responsible for creating it
system = aifutils.make_system_with_uri(
g, "http://www.test.edu/testSystem")
# it doesn't matter what URI we give entities, events, etc. so long as they are
# unique
entity = aifutils.make_entity(g, "http://www.test.edu/entities/1",
system)
# in order to allow uncertainty about the type of an entity, we don't mark an
# entity's type directly on the entity, but rather make a separate assertion for it
# its URI doesn't matter either
type_assertion = aifutils.mark_type(
g, "http://www.test.org/assertions/1", entity,
SEEDLING_TYPES_NIST.Person, system, 1.0)
# the justification provides the evidence for our claim about the entity's type
# we attach this justification to both the type assertion and the entity object
# itself, since it provides evidence both for the entity's existence and its type.
# in TA1 -> TA2 communications, we attach confidences at the level of justifications
aifutils.mark_text_justification(g, [entity, type_assertion],
"NYT_ENG_20181231", 42, 143, system,
0.973)
# let's suppose we also have evidence from an image
bb1 = Bounding_Box((123, 45), (167, 98))
aifutils.mark_image_justification(g, [entity, type_assertion],
"NYT_ENG_20181231_03", bb1, system,
0.123)
# and also a video where the entity appears in a keyframe
bb2 = Bounding_Box((123, 45), (167, 98))
aifutils.mark_keyframe_video_justification(g, [entity, type_assertion],
"NYT_ENG_20181231_03",
"keyframe ID", bb2, system,
0.234)
# and also a video where the entity does not appear in a keyframe
aifutils.mark_shot_video_justification(g, [entity, type_assertion],
"SOME_VIDEO", "some shot ID",
system, 0.487)
# and even audio!
aifutils.mark_audio_justification(g, [entity, type_assertion],
"NYT_ENG_201181231", 4.566, 9.876,
system, 0.789)
# also we can link this entity to something in an external KB
aifutils.link_to_external_kb(g, entity, "freebase.FOO", system, .398)
# let's mark our entity with some arbitrary system-private data. You can attach such data
# to nearly anything
aifutils.mark_private_data(g, entity, "{ 'hello' : 'world' }", system)
self.new_file(
g,
"test_create_an_entity_with_all_justification_types_and_confidence.ttl"
)
self.dump_graph(g, "Example of entity with all justifications")
def transferAIF(parent):
chi = child[parent]
if parent in kb_dict.keys():
g = kb_dict[parent]
else:
g = aifutils.make_graph()
entityDic_c = {}
country_set_r = set()
for img_id in chi:
if img_id in flag_dict_s.keys():
country_set_r.add(flag_dict_s[img_id])
#=============== Flag ===========
sys = aifutils.make_system_with_uri(
g, "http://www.columbia.edu/AIDA/DVMM/Systems/Flag/Inception_v4")
for key, value in index_category_f.items():
if key not in country_set_r:
continue
key = key.replace(' ', '_')
name = "http://www.columbia.edu/AIDA/DVMM/Entities/Country/" + str(
value) + '/' + key
entity = aifutils.make_entity(g, name, sys)
entityDic_c[key] = entity
#print "parent "+ parent
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/Country/"\
+str(value)+'/'+key, entity, AIDA_PROGRAM_ONTOLOGY_G.term('GPE'), sys, 1)
for img_id in chi:
if img_id in flag_dict_s.keys():
#print "imgID "+img_id
flag_label = [
'/m/07cmd', '/m/0dzct', '/m/03bt1vf', '/m/01g317', '/m/04yx4',
'/m/01prls', '/m/07yv9', '/m/03120'
]
key = img_id
if key in OD_result.keys():
for n in range(len(OD_result[key])):
eid = "http://www.columbia.edu/AIDA/DVMM/Entities/ObjectDetection/RUN00010/JPG/" + key + "/" + str(
n)
#print OD_result[key][n]['label']
if OD_result[key][n]['label'] in flag_label:
#print OD_result[key][n]['label']
eid = "http://www.columbia.edu/AIDA/DVMM/Entities/ObjectDetection/RUN00010/JPG/" + key + "/" + str(
n)
relation_entity = aifutils.make_relation(g, "http://www.columbia.edu/AIDA/DVMM/Relaion/Flag/"+\
key+"/"+str(n), sys)
#APORA = 'GeneralAffiliation.ArtifactPoliticalOrganizationReligiousAffiliation'
score = 1
type_assertion = aifutils.mark_type(g, "http://www.columbia.edu/AIDA/DVMM/TypeAssertion/Flag/"+key+"/"+str(n), \
relation_entity, AIDA_PROGRAM_ONTOLOGY_A['GeneralAffiliation.ArtifactPoliticalOrganizationReligiousAffiliation.NationalityCitizen'], \
sys, score)
boxA = OD_result[key][n]['bbox']
bb2 = Bounding_Box((boxA[0], boxA[1]),
(boxA[2], boxA[3]))
score = 1
justif = aifutils.mark_image_justification(g, [relation_entity, \
type_assertion], key, bb2, sys, score)
aifutils.add_source_document_to_justification(
g, justif, parent)
aifutils.mark_informative_justification(
g, relation_entity, justif)
score = 1
if eid in entity_dict.keys():
art_argument = aifutils.mark_as_argument(g, relation_entity, \
AIDA_PROGRAM_ONTOLOGY_A['GeneralAffiliation.ArtifactPoliticalOrganizationReligiousAffiliation.NationalityCitizen_Artifact'], \
entity_dict[eid], sys, score)
score = 1
justif = aifutils.mark_image_justification(g, [relation_entity, \
art_argument], key, bb2, sys, score)
aifutils.add_source_document_to_justification(
g, justif, parent)
aifutils.mark_informative_justification(
g, relation_entity, justif)
score = 1
nation_argument = aifutils.mark_as_argument(g, relation_entity, \
AIDA_PROGRAM_ONTOLOGY_A['GeneralAffiliation.ArtifactPoliticalOrganizationReligiousAffiliation.NationalityCitizen_Nationality'], \
entityDic_c[flag_dict_s[img_id]], \
sys, score)
score = 1
justif = aifutils.mark_image_justification(g, [relation_entity, \
nation_argument], key, bb2, sys, score)
aifutils.add_source_document_to_justification(
g, justif, parent)
aifutils.mark_informative_justification(
g, relation_entity, justif)
if OD_result[key][n]['label'] == '/m/03120':
#print flag_dict_s[img_id]
boxA = OD_result[key][n]['bbox']
bb2 = Bounding_Box((boxA[0], boxA[1]),
(boxA[2], boxA[3]))
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/Country/RUN00010/JPG/"+\
str(key)+'/'+str(n), entityDic_c[flag_dict_s[img_id]], AIDA_PROGRAM_ONTOLOGY_G.term('GPE'), sys, 1)
score = 1
justif = aifutils.mark_image_justification(g, [entityDic_c[flag_dict_s[img_id]], type_assertion], \
key, bb2, sys, score)
aifutils.add_source_document_to_justification(
g, justif, parent)
aifutils.mark_informative_justification(
g, entityDic_c[flag_dict_s[img_id]], justif)
#break
#=============== landmark ==========
sys = aifutils.make_system_with_uri(
g, "http://www.columbia.edu/AIDA/DVMM/Systems/Landmark/Delf")
#building
has_land = 0
landmark_name_set = set()
for imageN in chi:
if imageN in landmark_dict.keys():
name_lm = landmark_dict[imageN]
if name_lm == '':
continue
#print name_lm
if name_lm not in landmark_name_set:
landmark_name_set.add(name_lm)
name = "http://www.columbia.edu/AIDA/DVMM/Entities/Landmark/"+ \
str(landmark_id[name_lm])+"/"+name_lm
entity = aifutils.make_entity(g, name, sys)
entityDic_b[name_lm] = entity
#print AIDA_PROGRAM_ONTOLOGY2.term('Person')
score = 1
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/Landmark/"+ \
str(landmark_id[name_lm])+"/"+name_lm \
, entity, AIDA_PROGRAM_ONTOLOGY_b.term('FAC'), sys, score)
if 'Maidan' in name_lm:
aifutils.link_to_external_kb(g, entity,
"LDC2019E43:80000020", sys, 1)
elif 'United_States_Capitol' in name_lm:
aifutils.link_to_external_kb(g, entity,
"LDC2019E43:4140827", sys, 1)
else:
aifutils.link_to_external_kb(g, entity,
VIS_ONTOLOGY_b.term(name_lm),
sys, 1)
for imageN in chi:
total_key.add(imageN)
if imageN in landmark_dict.keys():
#print parent
key = imageN
name_lm = landmark_dict[imageN]
if name_lm == '':
continue
#aifutils.mark_as_possible_cluster_member(g, \
# entity,clusterDic_b[0], 1, sys)
# Need to change in future
im = Image.open(input_img_path + imageN + '.jpg')
width, height = im.size
bb2 = Bounding_Box((0, 0), (width, height)) #l u r d
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/Landmark/RUN00010/JPG/"+\
str(key)+'/'+str(0), entityDic_b[name_lm], AIDA_PROGRAM_ONTOLOGY_b.term('FAC'), sys, 1)
justif = aifutils.mark_image_justification(
g, [entityDic_b[name_lm], type_assertion], key, bb2, sys, 1)
aifutils.add_source_document_to_justification(g, justif, parent)
aifutils.mark_informative_justification(g, entityDic_b[name_lm],
justif)
#=============face recognition=============
sys = aifutils.make_system_with_uri(
g, "http://www.columbia.edu/AIDA/DVMM/Systems/Face/FaceNet")
NameDetected = set()
NameDetected_score = {}
nameCount = Counter()
count = 0
c_num = 0
person_c_n = 0
#print comblineSet
In = 0
#print bb
featureDic = {}
first = 1
entityList = []
arrayList = []
person_label = [
'/m/01g317', '/m/04yx4', '/m/03bt1vf', '/m/01bl7v', '/m/05r655',
'/m/04hgtk', '/m/01bgsw'
]
for x, y in result.items():
#print x
data = x.split('/')[-1]
if '._' in data:
continue
data2 = data.split('_')
#print data2[0]
#print data2[1][:-4]
key = data2[0]
i = data2[1][:-4]
#print chi
#print key
if key not in chi:
continue
if y[0].replace(' ', '_') not in nameSet or nameDic[y[0].replace(
' ', '_')] not in p1Set or float(y[1]) < 0.04 or y[0].replace(
' ', '_') == 'Ban_Ki-moon':
continue
else:
NameDetected.add(nameDic[y[0].replace(' ', '_')])
score = float(y[1]) + 0.5
NameDetected_score[nameDic[y[0].replace(' ', '_')]] = min(1, score)
for x, y in result2.items():
#print x
data = x.split('/')[-2]
if '._' in data:
continue
data = x.split('/')[-1]
data2 = data.split('_')
key = x.split('/')[-2]
if videoDic[key] not in chi:
continue
if y[0].replace(' ', '_') not in nameSet or nameDic[y[0].replace(
' ', '_')] not in p1Set or float(y[1]) < 0.04 or y[0].replace(
' ', '_') == 'Ban_Ki-moon':
continue
else:
NameDetected.add(nameDic[y[0].replace(' ', '_')])
score = float(y[1]) + 0.5
NameDetected_score[nameDic[y[0].replace(' ', '_')]] = min(1, score)
for key, value in index_category.items(): #key name. value is number
if key not in NameDetected:
continue
key = key.replace(' ', '_')
#keu = entityDic
name = "http://www.columbia.edu/AIDA/DVMM/Entities/FaceID/" + str(
value) + '/' + key
entity = aifutils.make_entity(g, name, sys)
entityDic[key] = entity
#print AIDA_PROGRAM_ONTOLOGY2.term('Person')
#score = NameDetected_score[key]
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/FaceID/"\
+str(value)+'/'+key, entity, AIDA_PROGRAM_ONTOLOGY2.term('PER'), sys, 1)
new_key = key.lower().replace('_', ' ')
if new_key in name2ID.keys():
aifutils.link_to_external_kb(g, entity,
"LDC2019E43:" + name2ID[new_key], sys,
1)
#print 'Lorelei'
#print parent
else:
a = 0
#print 'dbpedia'
#print parent
aifutils.link_to_external_kb(g, entity, VIS_ONTOLOGY2.term(key),
sys, 1)
for x, y in result.items():
#print x
data = x.split('/')[-1]
if '._' in data:
continue
data2 = data.split('_')
#print data2[0]
#print data2[1][:-4]
key = data2[0]
i = data2[1][:-4]
#print chi
#print key
if key not in chi:
continue
#print chi
#print key
In = 1
name = "http://www.columbia.edu/AIDA/DVMM/Entities/FaceDetection/RUN00010/" + str(
key) + '/' + str(i)
#entityList.append(key)
#entityList.append(entity)
entity = aifutils.make_entity(g, name, sys)
first_cluster = 1
#======================== JPG ===============
eid_list = []
if key in OD_result.keys():
for n in range(len(OD_result[key])):
#print OD_result[key][n]['label']
if OD_result[key][n]['label'] in person_label:
#print OD_result[key][n]['label']
boxA = OD_result[key][n]['bbox']
boxB = (int(bb[x][0]), int(bb[x][1]), int(bb[x][2]),
int(bb[x][3]))
IOA = bb_intersection_over_union(boxA, boxB)
if IOA > 0.9:
#left,top,right,bottom = OD_result[key][n]['bbox']
#if int(bb[x][1]) > left and int(bb[x][0]) > top and int(bb[x][3]) < right and int(bb[x][2]) < bottom:
eid = "http://www.columbia.edu/AIDA/DVMM/Entities/ObjectDetection/RUN00010/JPG/" + key + "/" + str(
n)
#print entity_dic2[key]
#print n
if n in entity_dic2[key]:
#if eid in entity_dict.keys():
score = IOA
eid_list.append(eid)
if first_cluster == 1:
first_cluster = 0
clusterName = aifutils.make_cluster_with_prototype(g, \
"http://www.columbia.edu/AIDA/DVMM/Clusters/HumanBody/RUN00010/JPG/"\
+key+"/"+str(n)+"/"+str(person_c_n),entity, sys)
aifutils.mark_as_possible_cluster_member(g, \
entity_dict[eid],clusterName, score, sys)
if first_cluster == 0:
person_c_n += 1
l, t, r, d = bb[x]
if (r - l) * (d - t) > 3600:
entityList.append(entity)
arrayList.append(y[2])
feature = {}
feature['columbia_vector_faceID_FaceNet'] = y[2].tolist()
json_data = json.dumps(feature)
aifutils.mark_private_data(g, entity, json_data, sys)
#labelrdf = VIS_ONTOLOGY.term(i_id)
#Dscore = value[i][7]
#if Dscore>1:
Dscore = 1
#type_assertion = aifutils.mark_type(g, "Columbia/DVMM/TypeAssertion/FaceRecognition/RUN00003/"+str(i_id)+"/"+str(i)+"/1",
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/FaceDetection/RUN00010/JPG/"+\
str(key)+'/'+str(i), entity, AIDA_PROGRAM_ONTOLOGY2.term('PER'), sys, Dscore)
bb2 = Bounding_Box((bb[x][0], bb[x][1]), (bb[x][2], bb[x][3]))
justif = aifutils.mark_image_justification(g, [entity, type_assertion],
key, bb2, sys, 1)
aifutils.add_source_document_to_justification(g, justif, parent)
aifutils.mark_informative_justification(g, entity, justif)
chi_set.add(key)
parent_set.add(parent)
if y[0].replace(' ', '_') not in nameSet or nameDic[y[0].replace(
' ', '_')] not in p1Set or float(y[1]) < 0.04 or y[0].replace(
' ', '_') == 'Ban_Ki-moon':
continue
else:
#nameCount2+=1
person_set.add(y[0])
doc_N.add(parent)
img_N.add(key)
score = sigmoid(float(y[1]) * 10)
NameDetected.add(nameDic[y[0].replace(' ', '_')])
#print y[0]
entity_key = nameDic[y[0].replace(' ', '_')]
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/FaceID/"\
+str(index_category[entity_key])+'/'+entity_key, entityDic[entity_key], AIDA_PROGRAM_ONTOLOGY2.term('PER'), sys, 1)
justif = aifutils.mark_image_justification(g, [entityDic[nameDic[y[0].replace(' ','_')]], \
type_assertion], key, bb2, sys, score)
aifutils.add_source_document_to_justification(g, justif, parent)
aifutils.mark_informative_justification(
g, entityDic[nameDic[y[0].replace(' ', '_')]], justif)
# entity, labelrdf, sys, score)
#print str(value[i][2]).replace("L",'')
#========================== Video Frame ================
for x, y in result2.items():
#print x
data = x.split('/')[-2]
if '._' in data:
continue
data = x.split('/')[-1]
data2 = data.split('_')
key = x.split('/')[-2]
if videoDic[key] not in chi:
continue
#print "video"
In = 1
i = data2[-1][:-4]
frame = data[:-len(data2[-1]) - 1]
frameNum = frame.split('_')[-1]
name = "http://www.columbia.edu/AIDA/DVMM/Entities/FaceDetection/RUN00010/Keyframe/"+\
str(videoDic[key])+'_'+str(frameNum)+'/'+str(i)
entity = aifutils.make_entity(g, name, sys)
if str(videoDic[key]) + '_' + str(frameNum) in ODF_result.keys():
first_cluster = 1
for n in range(
len(ODF_result[str(videoDic[key]) + '_' + str(frameNum)])):
if ODF_result[str(videoDic[key]) + '_' +
str(frameNum)][n]['label'] in person_label:
boxA = ODF_result[str(videoDic[key]) + '_' +
str(frameNum)][n]['bbox']
boxB = (int(bb_2[x][0]), int(bb_2[x][1]), int(bb_2[x][2]),
int(bb_2[x][3]))
IOA = bb_intersection_over_union(boxA, boxB)
if IOA > 0.9:
eid = "http://www.columbia.edu/AIDA/DVMM/Entities/ObjectDetection/RUN00010/Keyframe/" + str(
videoDic[key]) + '_' + str(frameNum) + "/" + str(n)
if n in entity_dic2[str(videoDic[key]) + '_' +
str(frameNum)]:
#if eid in entity_dict.keys():
score = IOA
#print x
#print entity_dict[eid]
#print n
if first_cluster == 1:
first_cluster = 0
clusterName = aifutils.make_cluster_with_prototype(g, \
"http://www.columbia.edu/AIDA/DVMM/Clusters/HumanBody/RUN00010/Keyframe/"+\
str(videoDic[key])+'_'+str(frameNum)+'/'+str(i)+'/'+\
str(person_c_n),entity, sys)
#aifutils.mark_as_possible_cluster_member(g, \
# entity,clusterName, score, sys)
aifutils.mark_as_possible_cluster_member(g, \
entity_dict[eid],clusterName, score, sys)
if first_cluster == 0:
person_c_n += 1
#txn.put("Columbia/DVMM/TypeAssertion/FaceID/RUN00003/"+str(key)+'/'+str(i), value[i][4]);
#featureDic[entity] = y[2]
featureDic[key] = y[2]
#entityList.append(key)
l, t, r, d = bb_2[x]
if (r - l) * (d - t) > 3600:
entityList.append(entity)
arrayList.append(y[2])
#if first == 1:
# new_array = [y[2]]
# first = 0
#else:
# new_array = np.concatenate((new_array, [y[2]]), axis=0)
feature = {}
feature['columbia_vector_faceID_FaceNet'] = y[2].tolist()
json_data = json.dumps(feature)
aifutils.mark_private_data(g, entity, json_data, sys)
#labelrdf = VIS_ONTOLOGY.term(i_id)
#Dscore = value[i][7]
#if Dscore>1:
Dscore = 1
#type_assertion = aifutils.mark_type(g, "Columbia/DVMM/TypeAssertion/FaceRecognition/RUN00003/"+str(i_id)+"/"+str(i)+"/1",
type_assertion = aifutils.mark_type(g, "http://www.columbia.edu/AIDA/DVMM/TypeAssertion/FaceDetection/RUN00010/Keyframe/"+\
str(videoDic[key])+'_'+str(frameNum)+'/'+str(i), entity, AIDA_PROGRAM_ONTOLOGY2.term('PER'), sys, Dscore)
#str(videoDic[key])+'_'+str(frameNum)+'/'+str(i), entity, AIDA_PROGRAM_ONTOLOGY2.Entity, sys, Dscore)
#print bb[x][1]
bb2 = Bounding_Box((bb_2[x][0], bb_2[x][1]), (bb_2[x][2], bb_2[x][3]))
#aifutils.mark_image_justification(g, [entity, type_assertion], key, bb2, sys, 1)
justif = aifutils.mark_keyframe_video_justification(g, [entity, type_assertion], videoDic[key], \
str(videoDic[key])+'_'+str(frameNum), bb2, sys, 1)
aifutils.add_source_document_to_justification(g, justif, parent)
aifutils.mark_informative_justification(g, entity, justif)
chi_set.add(key)
parent_set.add(parent)
if y[0].replace(' ', '_') not in nameSet or nameDic[y[0].replace(
' ', '_')] not in p1Set or float(y[1]) < 0.04 or y[0].replace(
' ', '_') == 'Ban_Ki-moon':
continue
else:
#nameCount2+=1
person_set.add(y[0])
doc_N.add(parent)
img_N.add(key)
#if float(y[1])*10>1:
# score = 1-random.random()/10
#else:
# score = float(y[1])*10
score = sigmoid(float(y[1]) * 10)
#place_of_birth_in_louisville_cluster = aifutils.mark_as_possible_cluster_member(g, \
# entity,clusterDic[nameDic[y[0].replace(' ','_')]], score, sys)
NameDetected.add(nameDic[y[0].replace(' ', '_')])
entity_key = nameDic[y[0].replace(' ', '_')]
type_assertion = aifutils.mark_type(g, \
"http://www.columbia.edu/AIDA/DVMM/TypeAssertion/FaceID/"\
+str(index_category[entity_key])+'/'+entity_key, entityDic[entity_key], AIDA_PROGRAM_ONTOLOGY2.term('PER'), sys, 1)
justif = aifutils.mark_keyframe_video_justification(g, [entityDic[nameDic[y[0].replace(' ','_')]], type_assertion], videoDic[key], \
str(videoDic[key])+'_'+str(frameNum), bb2, sys, score)
aifutils.add_source_document_to_justification(g, justif, parent)
aifutils.mark_informative_justification(
g, entityDic[nameDic[y[0].replace(' ', '_')]], justif)
#dbscan_run(arrayList,entityList)
new_array = np.array(arrayList)
if len(arrayList) > 1:
# Compute DBSCAN
#if __name__ == '__main__':
db = DBSCAN(eps=0.55, min_samples=2).fit(new_array)
core_samples_mask = np.zeros_like(db.labels_, dtype=bool)
core_samples_mask[db.core_sample_indices_] = True
labels = db.labels_
#print labels
# Number of clusters in labels, ignoring noise if present.
n_clusters_ = len(set(labels)) - (1 if -1 in labels else 0)
#print entityList
#print('Estimated number of clusters: %d' % n_clusters_)
clusterNameDic = {}
firstMem = [0 for i in range(n_clusters_)]
firstArray = {}
for i in range(len(labels)):
if labels[i] == -1:
continue
#print len(labels)
#print len(entityList)
#score = 1
if firstMem[labels[i]] == 0:
firstMem[labels[i]] = 1
firstArray[labels[i]] = new_array[i]
clusterNameDic[labels[i]] = aifutils.make_cluster_with_prototype(g, \
"http://www.columbia.edu/AIDA/DVMM/Clusters/FaceCoreference/RUN00010/"+\
str(labels[i]),entityList[i], sys)
#print entityList[a][j]
else:
dist = np.linalg.norm(firstArray[labels[i]] - new_array[i])
if dist > 1:
score = 0.001
else:
score = 1 - dist / 2
#score = sigmoid(dist)
#print score
aifutils.mark_as_possible_cluster_member(g, \
entityList[i],clusterNameDic[labels[i]], score, sys)
sys = aifutils.make_system_with_uri(
g, "http://www.columbia.edu/AIDA/DVMM/Systems/Face/FaceNet")
for key, value in index_category.items():
if key not in NameDetected:
continue
key = key.replace(' ', '_')
#print name2ID[]
new_key = key.lower().replace('_', ' ')
#print new_key
try:
#print RPI[parent].keys()
#print name2ID[new_key]
if name2ID[new_key] in RPI[parent].keys():
print(new_key)
#print parent
cluster = aifutils.make_cluster_with_prototype(g, \
"http://www.columbia.edu/AIDA/DVMM/Clusters/NamedPersonCoreference/"+\
str(value)+'/'+key,entityDic[key], sys)
score = 1
#aifutils.mark_as_possible_cluster_member(g, ,cluster, score, sys)
for i in range(len(RPI[parent][name2ID[new_key]])):
aifutils.mark_as_possible_cluster_member(g, \
RPI[parent][name2ID[new_key]][i],cluster, score, sys)
except KeyError:
a = 0
directory = ttl_out + '/'
if not os.path.exists(directory):
os.makedirs(directory)
with open(directory + parent + '.ttl', 'w') as fout:
serialization = BytesIO()
# need .buffer because serialize will write bytes, not str
g.serialize(destination=serialization, format='turtle')
fout.write(serialization.getvalue().decode('utf-8'))
def add_detections_to_graph(g, detections, parent_id, imgid, is_keyframe):
str_append = 'Keyframe' if is_keyframe else 'JPG'
if is_keyframe:
vidid = imgid.split('_')[0]
for ii, det in enumerate(detections):
label = det['label']
score = det['score']
bbox = det['bbox']
model = det['model']
if score < score_threshold:
continue
if model.startswith('coco'):
sys = system_co
elif model.startswith('voc'):
sys = system_pa
elif model.startswith('oi'):
sys = system_oi
elif model.startswith('ws'):
sys = system_ws
else:
raise
for iii, ont_id in enumerate(class2ont[label]):
ont_name = ont2name[ont_id]
labelrdf = getattr(ldc_ontology_m36, '_'.join(ont_name.split('.')),
None)
if labelrdf is None:
raise ValueError(ont_name)
if ont_id.startswith('LDC_ent_'):
eid = f"http://www.columbia.edu/AIDA/DVMM/Entities/ObjectDetection/RUN00010/{str_append}/{imgid}/{ii}"
entity = aifutils.make_entity(g, eid, sys)
entity_dict[eid] = entity
type_assertion = aifutils.mark_type(
g,
eid.replace('Entities', 'TypeAssertions') + f'/{iii}',
entity, labelrdf, sys, score)
bb = Bounding_Box((bbox[0], bbox[1]), (bbox[2], bbox[3]))
if is_keyframe:
justif = aifutils.mark_keyframe_video_justification(
g, [entity, type_assertion], vidid, imgid, bb, sys,
score)
else:
justif = aifutils.mark_image_justification(
g, [entity, type_assertion], imgid, bb, sys, score)
aifutils.add_source_document_to_justification(
g, justif, parent_id)
aifutils.mark_informative_justification(g, entity, justif)
#aifutils.mark_private_data(g, entity, json.dumps({}), sys)
#if ont_name.split('.')[0] in allowed_to_have_name:
# aifutils.mark_name(g, entity, class_names[label].split('(')[-1][:-1])
elif ont_id.startswith('LDC_evt_'):
eid = f"http://www.columbia.edu/AIDA/DVMM/Events/ObjectDetection/RUN00010/{str_append}/{imgid}/{ii}"
event = aifutils.make_event(g, eid, sys)
event_dict[eid] = event
type_assertion = aifutils.mark_type(
g,
eid.replace('Events', 'TypeAssertions') + f'/{iii}', event,
labelrdf, sys, score)
bb = Bounding_Box((1, 1), image_shape[imgid])
if is_keyframe:
justif = aifutils.mark_keyframe_video_justification(
g, [event, type_assertion], vidid, imgid, bb, sys,
score)
else:
justif = aifutils.mark_image_justification(
g, [event, type_assertion], imgid, bb, sys, score)
aifutils.add_source_document_to_justification(
g, justif, parent_id)
aifutils.mark_informative_justification(g, event, justif)
#aifutils.mark_private_data(g, event, json.dumps({}), sys)
#aifutils.mark_name(g, event, class_names[label])
else:
raise NotImplementedError