def runImageSimilaritySearch(self, params):
assert hasattr(self, 'search_url')
classifications = json.loads(
params['classifications']) if 'classifications' in params else []
params['n'] = params['n'] if 'n' in params else str(DEFAULT_PAGE_SIZE)
image, _type = base64FromUrl(params['url'])
smqtk_r = requests.get(self.search_url + '/n=' + params['n'] +
'/base64://' + image + '?content_type=' + _type)
assert smqtk_r.ok
smqtk_r = smqtk_r.json()
neighbors_to_distances = dict(
zip(smqtk_r['neighbors'], smqtk_r['distances']))
documents = solr_documents_from_field('sha1sum_s_md',
neighbors_to_distances.keys(),
classifications)
for document in documents:
document['smqtk_distance'] = neighbors_to_distances[
document['sha1sum_s_md']]
if 'near_duplicates' in params and int(params['near_duplicates']) == 1:
documents = [
x for x in documents
if x['smqtk_distance'] <= NEAR_DUPLICATES_THRESHOLD
]
documents = sorted(
documents, key=lambda x: x['smqtk_distance'])[:int(params['n'])]
return {'numFound': len(documents), 'docs': documents}
def results(self, params):
offset = int(params['offset'] if 'offset' in params else 0)
limit = int(params['limit'] if 'limit' in params else 20)
resp = requests.get(self.search_url + '/get_results', params={
'sid': params['sid'],
'i': offset,
'j': offset + limit
}).json() # @todo handle errors
documents = solr_documents_from_field('sha1sum_s_md', [sha for (sha, _) in resp['results']])
# The documents from Solr (since shas map to >= 1 document) may not be in the order of confidence
# returned by IQR, sort the documents to match the confidence values.
# Sort by confidence values first, then sha checksums second so duplicate images are grouped together
confidenceValues = dict(resp['results']) # Mapping of sha -> confidence values
if len(documents) < len(resp['results']):
logger.error('SID %s: There are SMQTK descriptors that have no corresponding Solr document(s).' % params['sid'])
for document in documents:
document['smqtk_iqr_confidence'] = confidenceValues[document['sha1sum_s_md']]
return {
'numFound': resp['total_results'],
'docs': sorted(documents,
key=lambda x: (x['smqtk_iqr_confidence'],
x['sha1sum_s_md']),
reverse=True)
}
def _search(self, params):
assert hasattr(self, 'search_url')
classifications = json.loads(
params['classifications']) if 'classifications' in params else []
cmu_images = requests.post(self.search_url,
data=params['url'],
headers={
'Content-type': 'text',
'Content-length':
str(len(params['url']))
},
verify=False).json()
cmu_images = [[
image.replace(setting.get('IMAGE_SPACE_CMU_PREFIX'),
setting.get('IMAGE_SPACE_SOLR_PREFIX')), score
] for (image, score) in cmu_images]
cmu_scores = {image.lower(): score for image, score in cmu_images}
documents = solr_documents_from_field('resourcename_t_md',
[x[0] for x in cmu_images],
classifications)
# Augment original scores from response into solr documents
for document in documents:
document['im_score'] = cmu_scores[document['id'].lower()]
return {'numFound': len(documents), 'docs': documents}
def _search(self, params):
assert hasattr(self, 'search_url')
classifications = json.loads(params['classifications']) if 'classifications' in params else []
cmu_images = requests.post(self.search_url,
data=params['url'],
headers={
'Content-type': 'text',
'Content-length': str(len(params['url']))
},
verify=False).json()
cmu_images = [[image.replace(os.environ['IMAGE_SPACE_CMU_PREFIX'],
os.environ['IMAGE_SPACE_SOLR_PREFIX']), score]
for (image, score) in cmu_images]
cmu_scores = {image.lower(): score for image, score in cmu_images}
documents = solr_documents_from_field('resourcename_t_md', [x[0] for x in cmu_images], classifications)
# Augment original scores from response into solr documents
for document in documents:
document['im_score'] = cmu_scores[document['id'].lower()]
return {
'numFound': len(documents),
'docs': documents
}
def runImageSimilaritySearch(self, params):
assert hasattr(self, 'search_url')
classifications = json.loads(params['classifications']) if 'classifications' in params else []
params['n'] = params['n'] if 'n' in params else str(DEFAULT_PAGE_SIZE)
smqtk_r = requests.get(self.search_url + '/n=' + params['n'] + '/' + params['url']).json()
neighbors_to_distances = dict(zip(smqtk_r['neighbors'], smqtk_r['distances']))
documents = solr_documents_from_field('sha1sum_s_md',
neighbors_to_distances.keys(),
classifications)
for document in documents:
document['smqtk_distance'] = neighbors_to_distances[document['sha1sum_s_md']]
if 'near_duplicates' in params and int(params['near_duplicates']) == 1:
documents = [x for x in documents if x['smqtk_distance'] <= NEAR_DUPLICATES_THRESHOLD]
return {
'numFound': len(documents),
'docs': documents
}
def results(self, params):
offset = int(params['offset'] if 'offset' in params else 0)
limit = int(params['limit'] if 'limit' in params else 20)
resp = requests.get(self.search_url + '/get_results',
params={
'sid': params['sid'],
'i': offset,
'j': offset + limit
}).json() # @todo handle errors
documents = solr_documents_from_field(
'sha1sum_s_md', [sha for (sha, _) in resp['results']])
# The documents from Solr (since shas map to >= 1 document) may not be in the order of confidence
# returned by IQR, sort the documents to match the confidence values.
# Sort by confidence values first, then sha checksums second so duplicate images are grouped together
confidenceValues = dict(
resp['results']) # Mapping of sha -> confidence values
if len(documents) < len(resp['results']):
logger.error(
'SID %s: There are SMQTK descriptors that have no corresponding Solr document(s).'
% params['sid'])
for document in documents:
document['smqtk_iqr_confidence'] = confidenceValues[
document['sha1sum_s_md']]
return {
'numFound':
resp['total_results'],
'docs':
sorted(documents,
key=lambda x:
(x['smqtk_iqr_confidence'], x['sha1sum_s_md']),
reverse=True)
}
def results(self, params):
def sid_exists(sid):
"""
Determine if a session ID already exists in SMQTK.
This currently creates the session if it doesn't already exist.
"""
return not requests.post(self.search_url + '/session',
data={
'sid': params['sid']
}).ok
offset = int(params['offset'] if 'offset' in params else 0)
limit = int(params['limit'] if 'limit' in params else 20)
if not sid_exists(params['sid']):
# Get pos/neg uuids from current session
session = self.model('item').findOne({'meta.sid': params['sid']})
if session:
self._refine({
'sid': params['sid'],
'pos_uuids': session['meta']['pos_uuids'],
'neg_uuids': session['meta']['neg_uuids']
})
resp = requests.get(self.search_url + '/get_results',
params={
'sid': params['sid'],
'i': offset,
'j': offset + limit
}).json() # @todo handle errors
try:
documents = solr_documents_from_field(
'sha1sum_s_md', [sha for (sha, _) in resp['results']])
except KeyError:
return {'numFound': 0, 'docs': []}
# The documents from Solr (since shas map to >= 1 document) may not be in the order of confidence
# returned by IQR, sort the documents to match the confidence values.
# Sort by confidence values first, then sha checksums second so duplicate images are grouped together
confidenceValues = dict(
resp['results']) # Mapping of sha -> confidence values
if len(documents) < len(resp['results']):
logger.error(
'SID %s: There are SMQTK descriptors that have no corresponding Solr document(s).'
% params['sid'])
for document in documents:
document['smqtk_iqr_confidence'] = confidenceValues[
document['sha1sum_s_md']]
return {
'numFound':
resp['total_results'],
'docs':
sorted(documents,
key=lambda x:
(x['smqtk_iqr_confidence'], x['sha1sum_s_md']),
reverse=True)
}