def overlaps(frame1, labels1, frame2, labels2):
"""
figures out which label of self matches which label of other
Assumes the zero label does not exist (background)
Returns sparse array of:
label in self (row)
label in other (col)
number of shared pixels (data)
"""
ki = np.empty(frame1.nnz, 'i')
kj = np.empty(frame2.nnz, 'i')
npx = cImageD11.sparse_overlaps(frame1.row, frame1.col, ki, frame2.row,
frame2.col, kj)
# self.data and other.data filled during init
row = frame1.pixels[labels1][ki[:npx]] # my labels
col = frame2.pixels[labels2][kj[:npx]] # your labels
ect = np.empty(npx, 'i') # ect = counts of overlaps
tj = np.empty(npx, 'i') # tj = temporary for sorting
n1 = frame1.meta[labels1]["nlabel"]
n2 = frame2.meta[labels2]["nlabel"]
tmp = np.empty(max(n1, n2) + 1, 'i') # for histogram
nedge = cImageD11.compress_duplicates(row, col, ect, tj, tmp)
# overwrites row/col in place : ignore the zero label (hope it is not there)
crow = row[:nedge] - 1
ccol = col[:nedge] - 1
cdata = ect[:nedge]
cedges = scipy.sparse.coo_matrix((cdata, (crow, ccol)), shape=(n1, n2))
# really?
return cedges
def overlaps(frame1, labels1, frame2, labels2):
"""
figures out which label of self matches which label of other
Returns sparse array of:
label in self (row)
label in other (col)
number of shared pixels (data)
"""
ki = np.empty(frame1.nnz, 'i')
kj = np.empty(frame2.nnz, 'i')
npx = cImageD11.sparse_overlaps(frame1.row, frame1.col, ki, frame2.row,
frame2.col, kj)
# self.data and other.data filled during init
row = frame1.pixels[labels1].data[ki[:npx]] # my labels
col = frame2.pixels[labels2].data[kj[:npx]] # your labels
ect = np.empty(npx, 'i') # ect = counts of overlaps
tj = np.empty(npx, 'i') # tj = temporary for sorting
n1 = frame1.pixels[labels1].header["nlabel"]
n2 = frame2.pixels[labels2].header["nlabel"]
tmp = np.empty(max(n1, n2) + 1, 'i') # for histogram
nedge = cImageD11.compress_duplicates(row, col, ect, tj, tmp)
# overwrites row/col in place
crow = row[:nedge]
ccol = col[:nedge]
cdata = ect[:nedge]
cedges = scipy.sparse.coo_matrix((cdata, (crow, ccol)))
return cedges
def overlaps(self, other):
"""
figures out which label of self matches which label of other
Returns sparse array of:
label in self (row)
label in other (col)
number of shared pixels (data)
"""
ki = np.empty(self.frame.nnz, 'i')
kj = np.empty(other.frame.nnz, 'i')
npx = cImageD11.sparse_overlaps(self.frame.row, self.frame.col, ki,
other.frame.row, other.frame.col, kj)
# self.data and other.data filled during init
row = self.labels[ki[:npx]] # my labels
col = other.labels[kj[:npx]] # your labels
ect = np.empty(npx, 'i') # ect = counts of overlaps
tj = np.empty(npx, 'i') # tj = temporary for sorting
tmp = np.empty(max(self.nlabel, other.nlabel) + 1,
'i') # for histogram
nedge = cImageD11.compress_duplicates(row, col, ect, tj, tmp)
# overwrites row/col in place
crow = row[:nedge]
ccol = col[:nedge]
cdata = ect[:nedge]
cedges = scipy.sparse.coo_matrix((cdata, (crow, ccol)))
return cedges
def test1(self):
np.random.seed(42)
N = self.N
a = np.random.random(N * 2)
i, j = (a * 10).astype('i').reshape(2, N)
oj = np.empty(N, 'i')
oi = np.empty(N, 'i')
tmp = np.empty(20, 'i')
d = np.ones(N, int)
r, c, d = _sum_duplicates(i, j, d)
n = cImageD11.compress_duplicates(i, j, oi, oj, tmp)
self.assertTrue(n == len(r))
self.assertTrue((r == i[:n]).all())
self.assertTrue((c == j[:n]).all())
self.assertTrue((d == oi[:n]).all())
def test2(self):
np.random.seed(42)
N = self.N
a = np.random.random(N * 2)
i, j = (a * 10).astype('i').reshape(2, N)
oj = np.empty(N, 'i')
oi = np.empty(N, 'i')
tmp = np.empty(20, 'i')
d = np.ones(N, int)
coo = scipy.sparse.coo_matrix((d, (i.copy(), j.copy())))
csr = coo.tocsr()
coo = csr.tocoo()
#print("",i,"\n",j)
n = cImageD11.compress_duplicates(i, j, oi, oj, tmp)
self.assertEqual(
cImageD11.sparse_is_sorted(i[:n].astype(np.uint16),
j[:n].astype(np.uint16)), 0)
r = coo.row
c = coo.col
d = coo.data
self.assertTrue(n == len(r))
self.assertTrue((r == i[:n]).all())
self.assertTrue((c == j[:n]).all())
self.assertTrue((d == oi[:n]).all())