def update_hash_templateL0(self, data):
knn = 3
if len(self._etalon) == 0:
self._etalon = data['clusters']
else:
matcher = Matcher(matcherForDetector(self.kodsettings.detector_type))
for index, et_cluster in enumerate(self._etalon):
dt_cluster = data['clusters'][index]
if et_cluster is None or len(et_cluster) == 0 or len(et_cluster) < knn:
self._etalon[index] = et_cluster
elif dt_cluster is None or len(dt_cluster) == 0 or len(dt_cluster) < knn:
self._etalon[index] = et_cluster
else:
dtype = dtypeForDetector(self.kodsettings.detector_type)
matches1 = matcher.knnMatch(listToNumpy_ndarray(et_cluster, dtype),
listToNumpy_ndarray(dt_cluster, dtype), k=knn)
matches2 = matcher.knnMatch(listToNumpy_ndarray(dt_cluster, dtype),
listToNumpy_ndarray(et_cluster, dtype), k=knn)
good = list(itertools.chain.from_iterable(itertools.imap(
lambda(x, _): (et_cluster[x.queryIdx], dt_cluster[x.trainIdx]), itertools.ifilter(
lambda(m, n): m.queryIdx == n.trainIdx and m.trainIdx == n.queryIdx, itertools.product(
itertools.chain(*matches1), itertools.chain(*matches2)
)
)
)))
self._etalon[index] = listToNumpy_ndarray(good)
def verify_template_L1(self, data):
knn = 2
matcher = Matcher(matcherForDetector(self.kodsettings.detector_type))
count = 0
prob = 0
logger.algo_logger.debug("Image: " + data['path'])
logger.algo_logger.debug("Template size: ")
for index, et_weight_cluster in enumerate(self._etalon):
d, c = itertools.izip(
*itertools.ifilter(lambda (_, c): c > 0, et_weight_cluster))
et_cluster = list(d)
cluster_weight = sum(c)
dt_cluster = data['clusters'][index]
if et_cluster is None or dt_cluster is None or len(
et_cluster) < knn or len(dt_cluster) < knn:
continue
if len(et_cluster) > 0 and len(dt_cluster) > 0:
dtype = dtypeForDetector(self.kodsettings.detector_type)
matches1 = matcher.knnMatch(
listToNumpy_ndarray(et_cluster, dtype),
listToNumpy_ndarray(dt_cluster, dtype),
k=2)
matches2 = matcher.knnMatch(
listToNumpy_ndarray(dt_cluster, dtype),
listToNumpy_ndarray(et_cluster, dtype),
k=2)
ms = [
et_cluster[x.queryIdx] for x in itertools.ifilter(
lambda (m, n): m.queryIdx == n.trainIdx and m.trainIdx
== n.queryIdx,
itertools.product(itertools.chain(
*matches1), itertools.chain(*matches2)))
]
c_val = sum(
lambda (_, x): x[1],
itertools.ifilter(
lambda (m, n): numpy.array_equal(m, n[0]),
itertools.product(iter(ms), iter(et_weight_cluster))))
count += 1
val = (c_val / (1.0 * cluster_weight)) * 100
logger.algo_logger.debug("Cluster #" + str(index + 1) + ": " +
str(cluster_weight) + " Positive: " +
str(c_val) + " Probability: " +
str(val))
prob += val
else:
logger.algo_logger.debug("Cluster #" + str(index + 1) + ": " +
str(cluster_weight) + " Invalid.")
logger.algo_logger.debug("Probability: " + str((prob / (1.0 * count))))
return prob / (1.0 * count)
def verify_template_L0(self, data):
knn = 2
matcher = Matcher(matcherForDetector(self.kodsettings.detector_type))
prob = 0
logger.algo_logger.debug("Image: " + data['path'])
logger.algo_logger.debug("Template size: ")
summ = sum(itertools.imap(lambda x: len(x) if x is not None else 0, self._etalon))
for index, et_cluster in enumerate(self._etalon):
dt_cluster = data['clusters'][index]
if et_cluster is None or len(et_cluster) < knn:
logger.algo_logger.debug("Cluster #" + str(index + 1) + ": " + str(-1)
+ " Invalid. (Weight: 0)")
continue
if dt_cluster is None or len(dt_cluster) < knn:
logger.algo_logger.debug("Cluster #" + str(index + 1) + ": " + str(len(self._etalon[index]))
+ " Positive: 0 Probability: 0 (Weight: " +
str(len(et_cluster) / (1.0 * summ)) + ")")
continue
if len(et_cluster) > 0 and len(dt_cluster) > 0:
dtype = dtypeForDetector(self.kodsettings.detector_type)
matches1 = matcher.knnMatch(listToNumpy_ndarray(et_cluster, dtype),
listToNumpy_ndarray(dt_cluster, dtype), k=knn)
matches2 = matcher.knnMatch(listToNumpy_ndarray(dt_cluster, dtype),
listToNumpy_ndarray(et_cluster, dtype), k=knn)
ms = [
x for (x, _) in itertools.ifilter(
lambda(m, n): m.queryIdx == n.trainIdx and m.trainIdx == n.queryIdx, itertools.product(
itertools.chain(*matches1), itertools.chain(*matches2)
)
)
]
val = (len(ms) / (1.0 * len(et_cluster))) * 100
logger.algo_logger.debug("Cluster #" + str(index + 1) + ": " + str(len(et_cluster))
+ " Positive: " + str(len(ms)) + " Probability: " + str(val) +
" (Weight: " + str(len(et_cluster) / (1.0 * summ)) + ")")
prob += (len(et_cluster) / (1.0 * summ)) * val
else:
logger.algo_logger.debug("Cluster #" + str(index + 1) + ": " + str(len(et_cluster))
+ " Invalid.")
logger.algo_logger.debug("Probability: " + str(prob))
return prob
def update_hash_templateL1(self, data):
"""
max_weight = se*sum(i=1,k-1: 1+2*i) + k*(n-k)*se,
where
n - count of images,
k - count of identical matches, k <= n,
se - single estimate, I used se=1
:param data:
:return:
"""
knn = 5
if len(self._database) == 1:
self._etalon = [[] if cluster is None else [(desc, 1)
for desc in cluster]
for cluster in data['clusters']]
else:
matcher = Matcher(
matcherForDetector(self.kodsettings.detector_type))
for index, et_cluster in enumerate(self._etalon):
dt_cluster = data['clusters'][index]
if dt_cluster is None or len(
dt_cluster) == 0 or len(dt_cluster) < knn:
continue
for obj in self._database:
if data['path'] == obj['path']:
continue
ob_cluster = obj['clusters'][index]
if ob_cluster is None or len(
ob_cluster) == 0 or len(ob_cluster) < knn:
continue
dtype = dtypeForDetector(self.kodsettings.detector_type)
matches1 = matcher.knnMatch(
listToNumpy_ndarray(dt_cluster, dtype),
listToNumpy_ndarray(ob_cluster, dtype),
k=knn)
for v in matches1:
if len(v) >= 1:
if v[0].distance == 0:
best = v[0]
else:
best = max(
v,
key=(lambda m: next(
(c / (1.0 * m.distance)
for (d, c) in et_cluster
if numpy.array_equal(
d, ob_cluster[m.trainIdx])), -1)))
ob_is = False
dt_is = False
new_cluster = []
for d, c in et_cluster:
if numpy.array_equal(
d, ob_cluster[best.trainIdx]):
c += 1
ob_is = True
if numpy.array_equal(
d, dt_cluster[best.queryIdx]):
c += 1
dt_is = True
new_cluster.append((d, c))
if not ob_is:
new_cluster.append(
(ob_cluster[best.trainIdx], 1))
if not dt_is:
new_cluster.append(
(dt_cluster[best.queryIdx], 1))
et_cluster = new_cluster
self._etalon[index] = et_cluster