def __call__(self, img):
# choose the pixels that are != 0 and not masked
nnz = select(img, self.msk, self.row, self.col, self.val, cut=self.cut)
# copy these into a "sparse" format
s = sparseframe.sparse_frame(self.row[:nnz], self.col[:nnz], img.shape)
s.set_pixels("intensity", self.val[:nnz])
# label them according to the connected objects
sparseframe.sparse_connected_pixels(s,
threshold=self.cut,
data_name='intensity',
label_name='cp')
# only keep spots with more than 3 pixels ...
mom = sparseframe.sparse_moments(s,
intensity_name='intensity',
labels_name='cp')
npx = mom[:, cImageD11.s2D_1]
pxcounts = npx[s.pixels['cp'] - 1]
msk = pxcounts > self.pixels_in_spot
if msk.sum() == 0:
return None
r = s.mask(msk)
sparseframe.sparse_connected_pixels(r,
threshold=self.cut,
label_name='connectedpixels',
data_name='intensity')
return r
def peaksearch(self, frm):
# -> f.pixels['localmax']
sparseframe.sparse_localmax(frm)
# -> f.pixels['connectedpixels']
sparseframe.sparse_connected_pixels(frm,
threshold=np.finfo(np.float32).min)
# which localmax peaks overlap each other
adj = sparseframe.overlaps(frm, 'localmax', frm, 'connectedpixels')
r, c = self.adj2overlap(adj)
return frm, r, c
def choose_parallel(args):
"""reads a frame and sends back a sparse frame"""
h5name, address, frame_num = args
frm = get_dset(h5name, address)[frame_num]
if choose_parallel.cache is None:
# cache the mallocs on this function. Should be one per process
row = np.empty(OPTIONS.MASK.size, np.uint16)
col = np.empty(OPTIONS.MASK.size, np.uint16)
val = np.empty(OPTIONS.MASK.size, frm.dtype)
choose_parallel.cache = row, col, val
else:
row, col, val = choose_parallel.cache
nnz = select(frm, OPTIONS.MASK, row, col, val, OPTIONS.CUT)
if nnz == 0:
sf = None
else:
if nnz > OPTIONS.HOWMANY:
nnz = top_pixels(nnz, row, col, val, OPTIONS.HOWMANY,
OPTIONS.THRESHOLDS)
# Now get rid of the single pixel 'peaks'
# (for the mallocs, data is copied here)
s = sparseframe.sparse_frame(row[:nnz].copy(), col[:nnz].copy(),
frm.shape)
s.set_pixels("intensity", val[:nnz].copy())
if OPTIONS.PIXELS_IN_SPOT <= 1:
sf = s
else:
# label them according to the connected objects
sparseframe.sparse_connected_pixels(s,
threshold=OPTIONS.CUT,
data_name="intensity",
label_name="cp")
# only keep spots with more than 3 pixels ...
mom = sparseframe.sparse_moments(s,
intensity_name="intensity",
labels_name="cp")
npx = mom[:, cImageD11.s2D_1]
pxcounts = npx[s.pixels["cp"] - 1]
pxmsk = pxcounts >= OPTIONS.PIXELS_IN_SPOT
if pxmsk.sum() == 0:
sf = None
else:
sf = s.mask(pxmsk)
return frame_num, sf
def newpks(
hf,
scans=None,
npixels=1,
monitor=None,
monval=1e3,
):
""" do a peak search in a sparse frame series """
titles = 's_raw f_raw avg_intensity Number_of_pixels sum_intensity omega dty'.split(
)
c = {}
for name in titles:
c[name] = []
# using the python file open overcomes some threading crap
with h5py.File(open(hf, "rb"), 'r') as hin:
# scan numbers
if scans is None:
scans = list(hin['/'])
for scan in scans:
gin = hin[scan]
shape = gin.attrs['shape0'], gin.attrs['shape1']
# import pdb; pdb.set_trace()
omega = gin['measurement/rot'][:]
difty = gin['instrument/positioners/dty'][()]
row = gin['row'][()]
col = gin['col'][()]
sig = gin['intensity'][()]
if monitor is not None:
mon = monval / gin[monitor][()]
ipt = np.cumsum(gin['nnz'][:])
iprev = 0
for k, nnz in enumerate(gin['nnz'][()]):
inext = iprev + nnz
if nnz == 0:
continue
f = sparseframe.sparse_frame(row[iprev:inext],
col[iprev:inext], shape)
f.set_pixels("intensity", sig[iprev:inext])
sparseframe.sparse_connected_pixels(f, threshold=0.1)
pks = sparseframe.sparse_moments(f, "intensity",
"connectedpixels")
# sparseframe.sparse_localmax(f)
# pks = sparseframe.sparse_moments(f, "intensity", "localmax")
s, f, a, n = moments(pks)
m = n > npixels
c['s_raw'].append(s[m])
c['f_raw'].append(f[m])
if monitor is not None:
c['avg_intensity'].append(a[m] * mon[k])
c['sum_intensity'].append(a[m] * n[m] * mon[k])
else:
c['avg_intensity'].append(a[m])
c['sum_intensity'].append(a[m] * n[m])
c['Number_of_pixels'].append(n[m])
npk = m.sum()
c['omega'].append(np.full(npk, omega[k]))
c['dty'].append(np.full(npk, difty))
iprev = inext
for t in titles:
c[t] = np.concatenate(c[t])
return c