def parse_sparse3d_scn(data):
"""
A function to retrieve sparse tensor input from larcv::EventSparseTensor3D object
Returns the data in format to pass to SCN
Args:
array of larcv::EventSparseTensor3D
Return:
voxels - numpy array(int32) with shape (N,3) - coordinates
data - numpy array(float32) with shape (N,C) - pixel values/channels
"""
meta = None
output = []
np_voxels = None
for event_tensor3d in data:
num_point = event_tensor3d.as_vector().size()
if meta is None:
meta = event_tensor3d.meta()
np_voxels = np.empty(shape=(num_point, 3), dtype=np.int32)
larcv.fill_3d_voxels(event_tensor3d, np_voxels)
else:
assert meta == event_tensor3d.meta()
np_data = np.empty(shape=(num_point, 1), dtype=np.float32)
larcv.fill_3d_pcloud(event_tensor3d, np_data)
output.append(np_data)
return np_voxels, np.concatenate(output, axis=-1)
def parse_sparse3d(data):
"""
A function to retrieve sparse tensor from larcv::EventSparseTensor3D object
Args:
length 1 array of larcv::EventSparseTensor3D
Return:
a numpy array with the shape (N,4) where 4=3+1 represents (x,y,z) coordinate and stored pixel value.
"""
event_tensor3d = data[0]
num_point = event_tensor3d.as_vector().size()
np_data = np.empty(shape=(num_point, 4), dtype=np.float32)
larcv.fill_3d_pcloud(event_tensor3d, np_data)
return np_data
def parse_sparse3d(event_tensor3d):
"""
A function to retrieve sparse tensor from larcv::EventSparseTensor3D object
Args:
event_tensor3d (larcv::EventSparseTensor3D): larcv C++ object for a 3d sparse tensor object
Return:
a pair of numpy arrays (coords,value), where coords has shape (N,3)
representing 3D pixel coordinate and value (N,1) stores pixel values.
"""
num_point = event_tensor3d.as_vector().size()
np_voxels = np.zeros(shape=(num_point, 3), dtype=np.int32)
np_data = np.zeros(shape=(num_point, 1), dtype=np.float32)
larcv.fill_3d_voxels(event_tensor3d, np_voxels)
larcv.fill_3d_pcloud(event_tensor3d, np_data)
return np_voxels, np_data
def parse_sparse3d_scn(data):
"""
A function to retrieve sparse tensor input from larcv::EventSparseTensor3D object
Returns the data in format to pass to SCN
Args:
length 1 array of larcv::EventSparseTensor3D
Return:
voxels - numpy array(int32) with shape (N,3) - coordinates
data - numpy array(float32) with shape (N,1) - pixel value
"""
event_tensor3d = data[0]
num_point = event_tensor3d.as_vector().size()
np_voxels = np.empty(shape=(num_point, 3), dtype=np.int32)
np_data = np.empty(shape=(num_point, 1), dtype=np.float32)
larcv.fill_3d_voxels(event_tensor3d, np_voxels)
larcv.fill_3d_pcloud(event_tensor3d, np_data)
return np_voxels, np_data
def initialize(self):
from larcv import larcv
from ROOT import TChain
ch_data = TChain('sparse3d_%s_tree' % self._flags.DATA_KEY)
ch_label = TChain('sparse3d_%s_tree' % self._flags.LABEL_KEY)
for f in self._flags.INPUT_FILE:
ch_data.AddFile(f)
ch_label.AddFile(f)
self._num_entries = ch_data.GetEntries()
self._data = []
self._label = []
br_data, br_label = (None, None)
event_fraction = 1. / self.num_entries() * 100.
total_point = 0.
for i in range(self.num_entries()):
ch_data.GetEntry(i)
ch_label.GetEntry(i)
if br_data is None:
br_data = getattr(ch_data,
'sparse3d_%s_branch' % self._flags.DATA_KEY)
br_label = getattr(
ch_label, 'sparse3d_%s_branch' % self._flags.LABEL_KEY)
num_point = br_data.as_vector().size()
np_data = np.zeros(shape=(num_point, 4), dtype=np.float32)
np_label = np.zeros(shape=(num_point, 4), dtype=np.float32)
larcv.fill_3d_pcloud(br_data, np_data)
larcv.fill_3d_pcloud(br_label, np_label)
self._data.append(np.expand_dims(np_data, 0))
self._label.append(np.expand_dims(np_label, 0))
total_point += np_data.size
sys.stdout.write(
'Processed %d%% ... %d MB\r' %
(int(event_fraction * i), int(total_point * 4 * 2 / 1.e6)))
sys.stdout.flush()
sys.stdout.write('\n')
sys.stdout.flush()
def parse_sparse3d(data):
"""
A function to retrieve sparse tensor from larcv::EventSparseTensor3D object
Args:
array of larcv::EventSparseTensor3D (one per channel)
Return:
a numpy array with the shape (N,3+C) where 3+C represents
(x,y,z) coordinate and C stored pixel values (channels).
"""
meta = None
output = []
for event_tensor3d in data:
num_point = event_tensor3d.as_vector().size()
if meta is None:
meta = event_tensor3d.meta()
np_voxels = np.empty(shape=(num_point, 3), dtype=np.int32)
larcv.fill_3d_voxels(event_tensor3d, np_voxels)
output.append(np_voxels)
else:
assert meta == event_tensor3d.meta()
np_values = np.empty(shape=(num_point, 1), dtype=np.float32)
larcv.fill_3d_pcloud(event_tensor3d, np_values)
output.append(np_values)
return np.concatenate(output, axis=-1)
def initialize(self):
self._last_entry = -1
self._event_keys = []
self._metas = []
# configure the input
from larcv import larcv
from ROOT import TChain
ch_data = TChain('sparse3d_%s_tree' % self._flags.DATA_KEY)
ch_label = None
if self._flags.LABEL_KEY:
ch_label = TChain('sparse3d_%s_tree' % self._flags.LABEL_KEY)
for f in self._flags.INPUT_FILE:
ch_data.AddFile(f)
if ch_label: ch_label.AddFile(f)
self._data = []
self._label = []
br_data,br_label=(None,None)
event_fraction = 1./ch_data.GetEntries() * 100.
total_point = 0.
for i in range(ch_data.GetEntries()):
ch_data.GetEntry(i)
if ch_label: ch_label.GetEntry(i)
if br_data is None:
br_data = getattr(ch_data, 'sparse3d_%s_branch' % self._flags.DATA_KEY)
if ch_label: br_label = getattr(ch_label, 'sparse3d_%s_branch' % self._flags.LABEL_KEY)
num_point = br_data.as_vector().size()
if num_point < 256: continue
np_data = np.zeros(shape=(num_point,4),dtype=np.float32)
larcv.fill_3d_pcloud(br_data, np_data)
self._data.append(np_data)
self._event_keys.append((br_data.run(),br_data.subrun(),br_data.event()))
self._metas.append(larcv.Voxel3DMeta(br_data.meta()))
if ch_label:
np_label = np.zeros(shape=(num_point,1),dtype=np.float32)
larcv.fill_3d_pcloud(br_label, np_label)
np_label = np_label.reshape([num_point]) - 1.
self._label.append(np_label)
total_point += np_data.size
sys.stdout.write('Processed %d%% ... %d MB\r' % (int(event_fraction*i),int(total_point*4*2/1.e6)))
sys.stdout.flush()
sys.stdout.write('\n')
sys.stdout.flush()
self._num_channels = self._data[-1].shape[-1]
self._num_entries = len(self._data)
# Output
if self._flags.OUTPUT_FILE:
import tempfile
cfg = '''
IOManager: {
Verbosity: 2
Name: "IOManager"
IOMode: 1
OutFileName: "%s"
InputFiles: []
InputDirs: []
StoreOnlyType: []
StoreOnlyName: []
}
'''
cfg = cfg % self._flags.OUTPUT_FILE
cfg_file = tempfile.NamedTemporaryFile('w')
cfg_file.write(cfg)
cfg_file.flush()
self._fout = larcv.IOManager(cfg_file.name)
self._fout.initialize()
def process(input_data, output_data):
"""
input_data and output_data types should be larcv.SparseTensor3D.
input_data holds 3D segmentation label values per voxel, (0,1,2,3,4) = (HIP,MIP,Shower,Delta,Michel)
output_data should be empty (or will be overwritten)
return is None
This function will change some voxels of (Delta,Michel) into Shower.
Those remain unchanged are the primary ionization trajectories attached to (HIP,MIP) pixels.
"""
if input_data.as_vector().size() < 1:
output_data.set(input_data, input_data.meta())
return
#t0=time.time()
voxels = np.zeros(shape=[input_data.as_vector().size(), 3], dtype=np.int32)
values = np.zeros(shape=[input_data.as_vector().size(), 1],
dtype=np.float32)
#if debug>0: print('c0', time.time()-t0)
#t0=time.time()
from larcv import larcv
larcv.fill_3d_voxels(input_data, voxels)
larcv.fill_3d_pcloud(input_data, values)
#if debug>0: print('c1', time.time()-t0)
values = values.squeeze(axis=1)
#t0=time.time()
no_correction = len(np.where(values > 2)[0]) == 0
trivial_correction = len(np.where(values < 3)[0]) == 0
#if debug>0: print('c2', time.time()-t0)
# Nothing to correct, return
if no_correction:
output_data.set(input_data, input_data.meta())
return
# Only michel/delta ray, make them all shower
if trivial_correction:
values[:] = 2.
vs = larcv.as_tensor3d(voxels, values, input_data.meta(), -1.)
output_data.set(vs, input_data.meta())
return
# Reaching here means something to correct.
# DBSCAN
#t0=time.time()
others_idx, others_vox = dbscan(voxels, values, find_valmax=1.)
deltas_idx, deltas_vox = dbscan(voxels, values, find_val=3., min_samples=1)
michels_idx, michels_vox = dbscan(voxels,
values,
find_val=4.,
min_samples=1)
#if debug>0: print('c3', time.time()-t0)
#t0=time.time()
correlated_deltas = correlate(others_vox, deltas_vox)
#if debug>0: print('c4', time.time()-t0)
#t0=time.time()
correlated_michels = correlate(others_vox, michels_vox)
#if debug>0: print('c5', time.time()-t0)
#t0=time.time()
for i, correlation in enumerate(correlated_deltas):
if correlation > 0: continue
values[deltas_idx[i]] = 2.
for i, correlation in enumerate(correlated_michels):
if correlation > 0: continue
values[michels_idx[i]] = 2.
#if debug>0: print('c6', time.time()-t0)
vs = larcv.as_tensor3d(voxels, values, input_data.meta(), -1.)
output_data.set(vs, input_data.meta())
return