-
Notifications
You must be signed in to change notification settings - Fork 0
/
DataLoader3D.py
401 lines (343 loc) · 16.3 KB
/
DataLoader3D.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
import ROOT
import numpy as np
from larcv import larcv
from larlite import larlite
from array import *
from larlite import larutil
from ublarcvapp import ublarcvapp
from MiscFunctions import cropped_np, unravel_array, reravel_array, paste_target
from MiscFunctions import unflatten_pos, flatten_pos, make_steps_images
from VoxelFunctions import Voxelator
def unstack(a, axis = 0):
return [np.squeeze(e, axis) for e in np.split(a, a.shape[axis], axis = axis)]
class DataLoader3D:
def __init__(self, PARAMS, verbose=False, all_train = False, all_valid = False, LArVoxMode=False):
self.PARAMS = PARAMS
self.LArVoxMode = LArVoxMode
self.voxelator = Voxelator(self.PARAMS) if not self.LArVoxMode else Voxelator(self.PARAMS, "LARVOXNETMICROBOONE")
self.verbose = verbose
self.infile = None
if all_train:
self.infile = ROOT.TFile(PARAMS['INFILE_TRAIN'])
elif all_valid:
self.infile = ROOT.TFile(PARAMS['INFILE_VAL'])
else:
self.infile = ROOT.TFile(PARAMS['INFILE'])
self.intree = self.infile.Get("TrackWalker3DInput") if not self.LArVoxMode else self.infile.Get("TrackWalker3DVoxInput")
self.RAND_FLIP_INPUTS = PARAMS['RAND_FLIP_INPUT']
self.nentries = self.intree.GetEntries()
self.nentries_train = int(self.nentries*0.8)
self.nentries_val = self.nentries-self.nentries_train
self.nentry_val_buffer = self.nentries_train
self.current_train_entry = 0
self.current_val_entry = 0
if all_train:
self.nentries_train = self.nentries
self.nentries_val = 0
self.nentry_val_buffer = self.nentries_train
elif all_valid:
self.nentries_val = self.nentries
self.nentries_train = 0
self.nentry_val_buffer = self.nentries_train
print()
print("Total Events in File: ", self.nentries)
print("Total Events in Training: ", self.nentries_train)
print("Total Events in Validation: ", self.nentries_val)
print()
self.PDG_to_Part = {
2212:"PROTON",
2112:"NEUTRON",
211:"PIPLUS",
-211:"PIMINUS",
111:"PI0",
11:"ELECTRON",
-11:"POSITRON",
13:"MUON",
-13:"ANTIMUON",
}
def get_train_data(self, n_load):
dim = 2*self.PARAMS['PADDING']+1
start_entry = self.current_train_entry
end_entry = self.current_train_entry + n_load
if end_entry > self.nentries_train:
end_entry = self.nentries_train
print("Loading Train Entries ", start_entry, "to", end=" ")
training_data = []
nAdded = 0
i = start_entry-1
while nAdded < n_load:
i += 1
if i >= self.nentries_train:
i = -1
continue
featcrops_np_v, xyzShifts_np_v, voxelsteps_final_np_v, isOK = self.get_data(i)
if isOK == False:
continue
training_data.append((featcrops_np_v,xyzShifts_np_v,voxelsteps_final_np_v))
nAdded += 1
self.current_train_entry = i+1
print(self.current_train_entry)
if self.current_train_entry == self.nentries_train:
self.current_train_entry = 0
return training_data
def get_data(self, i):
self.intree.GetEntry(i)
# np array of x,y,z,StepIDX in full detector voxel coord
voxelsteps_np = self.intree.voxelsteps_np.tonumpy().copy()
# Min Row, Min Cols for the feature images (to offset vox projection)
minVox_np = self.intree.minVoxCoords_np.tonumpy().astype(np.int32).copy()
maxVox_np = self.intree.maxVoxCoords_np.tonumpy().astype(np.int32).copy()
# In (xVox,yVox,zVox,32 Feats) a sparse array of feats around the given track
sparse_feats_np = self.intree.feats_np.tonumpy().copy()
# if there is only 1 or fewer points on the index map then dont include track
if voxelsteps_np.shape[0] < 2 or len(sparse_feats_np.shape) != 2:
print("Skipping issue with Saved ROOT Track Shape")
return None, None, None, 0
featcrops_np_v, xyzShifts_np_v, voxelsteps_final_np_v = \
self.make_track_crops(sparse_feats_np, voxelsteps_np, minVox_np, maxVox_np, self.PARAMS)
if len(featcrops_np_v) < 2:
print("Skipping issue with Saved ROOT Track Shape after making track crops")
return None, None, None, 0
if self.RAND_FLIP_INPUTS:
print("Random flipping of images not implemented for complex dataloading")
assert 1==2
return featcrops_np_v, xyzShifts_np_v, voxelsteps_final_np_v, 1
def set_current_entry(self,entry):
self.current_train_entry = entry
self.current_val_entry = entry
def get_val_data(self, n_load):
dim = 2*self.PARAMS['PADDING']+1
start_entry = self.current_val_entry
end_entry = self.current_val_entry + n_load
if end_entry > self.nentries_val:
end_entry = self.nentries_val
self.current_val_entry = 0
start_entry = self.current_val_entry
end_entry = self.current_val_entry + n_load
# print("Loading Val Entries ", start_entry+self.nentry_val_buffer, "to", end_entry+self.nentry_val_buffer)
val_data = []
nAdded = 0
i = start_entry-1
while nAdded < n_load:
i += 1
if i >= self.nentries_val:
i = -1
continue
featcrops_np_v, xyzShifts_np_v, voxelsteps_final_np_v, isOK = self.get_data(i)
if isOK == False:
continue
val_data.append((featcrops_np_v,xyzShifts_np_v))
nAdded += 1
self.current_val_entry = i+1
if self.current_val_entry == self.nentries_val:
self.current_val_entry = 0
return val_data
def make_track_crops(self, sparse_feats_np, voxelsteps_np, minVox_np, maxVox_np, PARAMS):
# print("Printing True Track")
# for idx in range(voxelsteps_np.shape[0]):
# print(voxelsteps_np[idx,:])
feat_idx = sparse_feats_np[:,0:3].copy().astype(np.int32)
feat_steps_np_v = []
xyzShifts_np_v = []
noshiftVoxelIdx_np_v = []
shiftVoxelIdx_np_v = []
targVoxelIdx_np_v = []
flattened_positions_v = []
area_positions_v = []
nSteps = 0
# Get First and Last Position
startVoxelPosition = voxelsteps_np[0,:].copy()
endVoxelPosition = voxelsteps_np[-1,:].copy()
thisVoxelPosition = startVoxelPosition.copy()
shiftedthisVoxelPosition = thisVoxelPosition.copy()
# Repopulate the part of the detector we saved for this track:
denseFeatPartialDetector = np.zeros((maxVox_np[0]-minVox_np[0],maxVox_np[1]-minVox_np[1],maxVox_np[2]-minVox_np[2],PARAMS['NFEATS']))
denseFeatPartialDetector[feat_idx[:,0]-minVox_np[0],feat_idx[:,1]-minVox_np[1],feat_idx[:,2]-minVox_np[2]] = sparse_feats_np[:,3:]
nextStepIdx = 0
voxelsteps_np_idx = 0
ct = 0
while np.array_equal(shiftedthisVoxelPosition[0:3],endVoxelPosition[0:3]) != True:
ct +=1
noshiftVoxelIdx_np_v.append(shiftedthisVoxelPosition)
if PARAMS['DO_CROPSHIFT']:
# for i in range(3):
# shift_amt = PARAMS['CROPSHIFT_MAXAMT']
# delta = np.random.randint(-shift_amt, shift_amt+1)
# shiftedthisVoxelPosition[i] +=delta
shiftedthisVoxelPosition[np.random.randint(0,3)] += np.random.randint(-PARAMS['CROPSHIFT_MAXAMT'],PARAMS['CROPSHIFT_MAXAMT']+1)
cropMins = [int(shiftedthisVoxelPosition[i]-PARAMS['PADDING']-minVox_np[i]) for i in range(3)]
cropMaxes = [int(shiftedthisVoxelPosition[i]+PARAMS['PADDING']+1-minVox_np[i]) for i in range(3)]
currentFeatCrop = denseFeatPartialDetector[cropMins[0]:cropMaxes[0],cropMins[1]:cropMaxes[1],cropMins[2]:cropMaxes[2]].copy()
if currentFeatCrop.shape != (PARAMS['PADDING']*2+1,PARAMS['PADDING']*2+1,PARAMS['PADDING']*2+1,PARAMS['NFEATS']):
print()
print("Failure to Crop correct shape of features")
print(shiftedthisVoxelPosition)
print(currentFeatCrop.shape)
print(denseFeatPartialDetector.shape)
print(cropMins[0],cropMaxes[0],cropMins[1],cropMaxes[1],cropMins[2],cropMaxes[2])
assert 1==2
# next step should be:
# within cropMins and maxes,
# less than or equal to PARAMS['TARG_STEP_DIST']
# The highest value in the voxelsteps_np[:,3] 'rank' possible
currentStepDist = 0
nextVoxelStep = np.array([-1.,-1.,-1.,-1.])
for stepIdx in range(voxelsteps_np_idx,voxelsteps_np.shape[0]):
testStep = voxelsteps_np[stepIdx,:].copy()
dist = ((shiftedthisVoxelPosition[0] - testStep[0])**2 + (shiftedthisVoxelPosition[1] - testStep[1])**2 + (shiftedthisVoxelPosition[2] - testStep[2])**2)**0.5
# print(" ", testStep,end='')
if dist <= PARAMS['TARG_STEP_DIST'] and testStep[3] > nextVoxelStep[3]:
nextVoxelStep = testStep
nextStepIdx = stepIdx
# print(nextVoxelStep, "Next Step Set",end='')
# print()
xyzShift = nextVoxelStep[0:3].copy() - shiftedthisVoxelPosition[0:3].copy()
# print(xyzShift)
# if np.sum(xyzShift*xyzShift)**0.5 < 2:
# print()
# print("Stepping:",np.sum(xyzShift*xyzShift)**0.5)
# print("Starting Debug")
# print(" ",shiftedthisVoxelPosition,"Starting Position")
# currentStepDist = 0
# nextVoxelStep = np.array([-1.,-1.,-1.,-1.])
# for stepIdx in range(voxelsteps_np_idx,voxelsteps_np.shape[0]):
# testStep = voxelsteps_np[stepIdx,:].copy()
# dist = ((shiftedthisVoxelPosition[0] - testStep[0])**2 + (shiftedthisVoxelPosition[1] - testStep[1])**2 + (shiftedthisVoxelPosition[2] - testStep[2])**2)**0.5
# # print(" ", testStep,end='')
# if dist <= PARAMS['TARG_STEP_DIST'] and testStep[3] > nextVoxelStep[3]:
# nextVoxelStep = testStep
# nextStepIdx = stepIdx
# print(" ",testStep,"Test Step",dist, " Target Position Set")
# else:
# print(" ",testStep,"Test Step",dist)
# assert 1==2
xyzShift = xyzShift/PARAMS['CONVERT_OUT_TO_DIST']
feat_steps_np_v.append(currentFeatCrop)
xyzShifts_np_v.append(xyzShift)
targVoxelIdx_np_v.append(nextVoxelStep)
shiftVoxelIdx_np_v.append(shiftedthisVoxelPosition)
# Change current position to next position
shiftedthisVoxelPosition = nextVoxelStep
cropMins = [int(endVoxelPosition[i]-PARAMS['PADDING']-minVox_np[i]) for i in range(3)]
cropMaxes = [int(endVoxelPosition[i]+PARAMS['PADDING']+1-minVox_np[i]) for i in range(3)]
feat_steps_np_v.append(denseFeatPartialDetector[cropMins[0]:cropMaxes[0],cropMins[1]:cropMaxes[1],cropMins[2]:cropMaxes[2]].copy())
xyzShifts_np_v.append(np.zeros((3,)).astype(np.float32))
targVoxelIdx_np_v.append(endVoxelPosition)
shiftVoxelIdx_np_v.append(endVoxelPosition)
feat_steps_np_v = np.stack(feat_steps_np_v,axis=0)
try:
xyzShifts_np_v = np.stack(xyzShifts_np_v,axis=0)
except:
for xyz in xyzShifts_np_v:
print(xyz)
assert 1==2
# print("Printing Track Info")
# for idx in range(len(targVoxelIdx_np_v)):
# print(shiftVoxelIdx_np_v[idx], targVoxelIdx_np_v[idx], xyzShifts_np_v[idx]*PARAMS['CONVERT_OUT_TO_DIST'], np.sum(xyzShifts_np_v[idx]*xyzShifts_np_v[idx])**0.5*PARAMS['CONVERT_OUT_TO_DIST'])
# assert 1==2
return feat_steps_np_v, xyzShifts_np_v, shiftVoxelIdx_np_v
def getNextStepClass(self, nextStepIdx_v, dim, isEndpoint=False ):
if isEndpoint == True:
val = int((dim-1)/2)
return int(val*dim*dim + val*dim + val) #13, refers to center of cube for endpoint
else:
classVal = nextStepIdx_v[0]*dim*dim + nextStepIdx_v[1]*dim + nextStepIdx_v[2]
return int(classVal)
def flip_target_idx_xdim(in_targ_idx,PARAMS):
out_targ_idx = -999*np.ones(in_targ_idx.shape)
for i in range(in_targ_idx.shape[0]):
pos_2d = unflatten_pos(in_targ_idx[i], PARAMS['PADDING']*2+1)
new_pos2d = [2*PARAMS['PADDING'] - pos_2d[0],pos_2d[1]]
pos_1d = flatten_pos(new_pos2d[0],new_pos2d[1], PARAMS['PADDING']*2+1)
out_targ_idx[i] = pos_1d
return out_targ_idx
def flip_flat_area_positions_xdim(in_flat_area_positions, PARAMS):
out_flat_area_positions = []
dim = PARAMS['PADDING']*2+1
for i in range(len(in_flat_area_positions)):
this_flat_area_position = in_flat_area_positions[i]
area_position_2d = this_flat_area_position.reshape(dim,dim)
area_position_2d = np.flip(area_position_2d,axis=0)
flat_flip_area = area_position_2d.flatten()
out_flat_area_positions.append(flat_flip_area)
return out_flat_area_positions
def row_get(y, origin_y, origin_y_plus_height, pixel_height):
if ((y < origin_y) or (y >= origin_y_plus_height)):
print("Row out of range", y, origin_y, origin_y_plus_height)
assert 1==2
else:
return int((y-origin_y)/pixel_height)
def col_get(x, origin_x, origin_x_plus_width, pixel_width):
if ((x < origin_x) or (x >= origin_x_plus_width)):
print("Row out of range", x, origin_x, origin_x_plus_width)
assert 1==2
else:
return int((x-origin_x)/pixel_width)
def getprojectedpixel_hardcoded(x,y,z):
nplanes = 3
fracpixborder = 1.5
pixel_height = 6.0
pixel_width = 1.0
DriftVelocity = 0.1098
SamplingRate = 500.0
min_y = 2400.0
max_y = 8448.0
rows = 1008
origin_y = 2400.0
origin_y_plus_height = 8448.0
min_x = 0.0
max_x = 3456.0
origin_x = 0.0
origin_x_plus_width = 3456.0
cols = 3456
row_border = fracpixborder*pixel_height;
col_border = fracpixborder*pixel_width;
img_coords = [-1,-1,-1,-1]
tick = x/(DriftVelocity*SamplingRate*1.0e-3) + 3200.0;
if ( tick < min_y ):
if ( tick > min_y- row_border ):
# below min_y-border, out of image
img_coords[0] = rows-1 # note that tick axis and row indicies are in inverse order (same order in larcv2)
else:
# outside of image and border
img_coords[0] = -1
elif ( tick > max_y ):
if (tick < max_y+row_border):
# within upper border
img_coords[0] = 0;
else:
# outside of image and border
img_coords[0] = -1;
else:
# within the image
img_coords[0] = col_get(tick,origin_y,origin_y_plus_height,pixel_height);
# Columns
# xyz = [ x, y, z ]
xyz = array('d', [x,y,z])
# there is a corner where the V plane wire number causes an error
if ( (y>-117.0 and y<-116.0) and z<2.0 ):
xyz[1] = -116.0;
for p in range(nplanes):
wire = larutil.Geometry.GetME().WireCoordinate( xyz, p );
# get image coordinates
if ( wire<min_x ):
if ( wire>min_x-col_border ):
# within lower border
img_coords[p+1] = 0;
else:
img_coords[p+1] = -1;
elif ( wire>=max_x ):
if ( wire<max_x+col_border ):
# within border
img_coords[p+1] = cols-1
else:
# outside border
img_coords[p+1] = -1
else:
# inside image
img_coords[p+1] = col_get(wire,origin_x,origin_x_plus_width,pixel_width) #meta.col( wire );
# end of plane loop
# there is a corner where the V plane wire number causes an error
if ( y<-116.3 and z<2.0 and img_coords[1+1]==-1 ):
img_coords[1+1] = 0;
return img_coords