-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataprep.py
368 lines (275 loc) · 11.4 KB
/
dataprep.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
from glob import glob
import numpy as np
import cv2
from tqdm import tqdm
from scipy import spatial
import torch
from torch.utils.data import Dataset, DataLoader, sampler
import torchvision.datasets as dset
from torchvision import transforms
from skimage import draw, io, exposure, morphology,transform
from PIL import Image
from joblib import Parallel, delayed
import multiprocessing
import mysegmentation as myseg
class MyDataset(Dataset):
def __init__(self, path, pytorch=True):
super().__init__()
self.raw_files = sorted(glob(path+"*raw.png"))
self.anno_files = sorted(glob(path+"*mask.png"))
self.pytorch=True
#self.colors = colors
def check_pad(self,im):
dim = im.shape[0]
if dim%32!=0:
res = int(dim/32)
diff = int(((res+1)*32-dim)/2)
im = np.pad(im,diff,"reflect")
return im
def load_image(self,idx, invert=True):
raw_rgb = io.imread(self.raw_files[idx])
#if invert:
#raw_rgb = raw_rgb.transpose((2,0,1))
norm = (raw_rgb / np.iinfo(raw_rgb.dtype).max)
norm = self.check_pad(norm)
return torch.unsqueeze(torch.tensor(norm, dtype=torch.float32),0)
def load_mask(self, idx):
"""load mask from path"""
mask = io.imread(self.anno_files[idx])/100
mask = self.check_pad(mask)
return torch.unsqueeze(torch.tensor(mask, dtype=torch.int64),0)
def __getitem__(self, idx):
image = self.load_image(idx)
mask = self.load_mask(idx)
return image, mask
def __len__(self):
return len(self.raw_files)
def __repr__(self):
s = 'Dataset class with {} files'.format(self.__len__())
return s
def load_train_val(train_SET,validation_SET,batch_size=12):
DLEN = int(train_SET.__len__())
VALEN = int(validation_SET.__len__())
ratio = (DLEN-VALEN,VALEN)
DATA_TRAIN = DataLoader(train_SET , batch_size=batch_size, shuffle=True)
DATA_VALID = DataLoader(validation_SET, batch_size=batch_size, shuffle=True)
return DATA_TRAIN, DATA_VALID
def mCPU(func, var, n_jobs=40,verbose=10):
return Parallel(n_jobs=n_jobs, verbose=verbose)(delayed(func)(i) for i in var)
def get_ds(M,S):
"""returns closest points: of main island M and slave island s:
returns: closest points of M and S and their distance D"""
T = spatial.cKDTree(M)
d,p = T.query(S)
m = np.argmin(d)
return S[m], M[p[m]], d[m]
def connect_nearest_dots(mask):
"""I forgot, what this does"""
unis = np.unique(mask)[1:]
dlists = [np.argwhere(mask==i) for i in unis]
l1 = dlists[0]
lS = dlists[1:]
M,S,D = [],[],[]
for i in lS:
m,s,d = get_ds(l1,i)
M.append(m)
S.append(s)
D.append(d)
MIN = np.argmin(D)
MP = M[MIN]
SP = S[MIN]
X,Y = [MP[1],SP[1]],[MP[0],SP[0]]
return Y,X
def connect_nearest(MA):
"""I forgot, what this does"""
Y, X = connect_nearest_dots(MA)
MA2 = MA.copy()
MA2[MA2>0] = 1
line = draw.line(Y[0],X[0],Y[1],X[1])
MA2[line] = 1
return myseg.get_segments(MA2)[0]
def fill_gaps(IM,max_iter=100):
"""I forgot, what this does"""
MA = IM.copy()
for i in range(100):
unilen = len(np.unique(MA))
if unilen < 3:
break
MA = connect_nearest(MA)
return MA
def data_from_coco(im_path, an_path, idn=100):
coco = dset.CocoDetection(root = im_path,
annFile = an_path)
im_ids = list(sorted(coco.coco.imgs.keys()))
imidn = im_ids[idn]
im_info = coco.coco.loadImgs([imidn])
image = cv2.imread(im_path+im_info[0]["file_name"])[:,:,0].astype("int")
annos = coco.coco.loadAnns(coco.coco.getAnnIds(imidn))
masks = []#np.zeros_like(image)
for n,i in tqdm(enumerate(annos)):
ma = coco.coco.annToMask(i)
ma = myseg.get_segments(ma)[0]
ma = fill_gaps(ma)
masks.append(ma)
#mask[ma==1] = n+1
#### the following will sort cells by area. Removes overlap info
areas = [len(np.argwhere(i==1)) for i in masks]
asort = np.argsort(areas) #not finnished
return image, np.asarray(masks)[asort]
def flatten_mask(masks):
mask = np.zeros(masks.shape[1:])
for n,i in enumerate(masks):
mask[i==1] = n+1
return mask
################################################################################
################################ Augmentations ################################
################################################################################
###### extract crops
def get_crop_pars(mask, ind=1, pad = 1.2):
"""get tight fitted crops (if pad = 1)
input: n x m numpy array
output: [1,1,1,1] <- [top,left,height,width] coordinates of rectangle """
anps = np.argwhere(mask == ind)
ymin, xmin = anps.min(0)
ymax, xmax = anps.max(0)
yc , xc = anps.mean(0)
dx = xmax-xmin
dy = ymax-ymin
d = np.max([dy,dx])*pad
top = int(yc-d/2)
left = int(xc-d/2)
height = int(d)
width = int(d)
return top,left,height,width
def create_crops(mask,pad=1.2):
croplist = []
cells = np.unique(mask)[1:]
for i in cells:
croplist.append(get_crop_pars(mask,i,pad=pad))
return np.asarray(croplist)
def to_PIL(im):
return Image.fromarray(np.uint8(im))
def crop_resize(IM,CROPPAR,s=512):
im = to_PIL(IM)
resize = transforms.Resize(size=(s, s),interpolation=transforms.InterpolationMode.NEAREST_EXACT)
i, j, h, w = CROPPAR
im = TF.crop(im, i, j, h, w)
return np.asarray(resize(im))
def im_resize(IM,CROPPAR,s=512):
im = to_PIL(IM)
resize = transforms.Resize(size=(s, s),interpolation=transforms.InterpolationMode.BILINEAR)
i, j, h, w = CROPPAR
im = TF.crop(im, i, j, h, w)
return np.asarray(resize(im))
def image_crop_resize(im,c=200,size=512):
im0 = transform.resize(im[c:-c,c:-c],(size,size),preserve_range=True,order=3)
return im0
def mask_crop_resize(im,c=200,size=512):
def cr(ma,c,size):
return transform.resize(ma[c:-c,c:-c],(size,size),preserve_range=True,anti_aliasing=True)
im0 = np.stack([cr(i,c,size) for i in tqdm(im)])
return im0.astype(int)
######## Distortions
def create_dist_maps(size,amp=10):
ysize, xsize = size
"""creates distortion map to apply to image or mask"""
def rd(dxy):
return (np.random.rand())
src_cols = np.linspace(0, xsize, 20)
src_rows = np.linspace(0, ysize, 20)
dx = src_cols[1]
dy = src_rows[1]
src_rows, src_cols = np.meshgrid(src_rows, src_cols)
src = np.dstack([src_cols.flat, src_rows.flat])[0]
dst_rows = src[:, 1] - np.sin(np.linspace(0, 2*rd(dy)*np.pi, src.shape[0])) * amp
dst_cols = src[:, 0] - np.cos(np.linspace(0, 2*rd(dx)*np.pi, src.shape[0])) * amp
#dst_rows *= 1.5
#dst_rows -= 1.5 * 50
dst = np.vstack([dst_cols, dst_rows]).T
mins0 = np.argwhere(src[:,0]==0)
mins1 = np.argwhere(src[:,1]==0)
maxs0 = np.argwhere(src[:,0]==src.max())
maxs1 = np.argwhere(src[:,1]==src.max())
dst = np.asarray(dst)
dst[:,0][mins0] = 0
dst[:,1][mins1] = 0
dst[:,0][maxs0] = src.max()
dst[:,1][maxs1] = src.max()
return [src, dst]
def distortions(image, maps,radius=None,shift=20,strength=0.1):
def shift_left(xy):
xy[:, 0] += shift
return xy
if radius == None:
radius = image.shape[0]*2
src, dst = maps
tform = transform.PiecewiseAffineTransform()
tform.estimate(src, dst)
dist = transform.warp(image, tform, output_shape=image.shape,
preserve_range=True,mode='reflect')
dist = transform.swirl(dist, rotation=0, strength=strength,
radius=radius,preserve_range=True,mode='reflect')
return dist
def distort_all(image,masks,amp=10,radius=2.,shift=20,strength=0.1):
dmap = create_dist_maps(image.shape,amp=10)
dimage = distortions(image, dmap,radius=radius,shift=shift,strength=strength)
def distort(mask):
dtemp = distortions(mask, dmap,radius=radius,shift=shift,strength=strength)
return np.round(dtemp,0).astype(int)
cpus = multiprocessing.cpu_count()
dmasks = np.stack(mCPU(distort,masks,cpus)).astype(int)
#dmask = flatten_mask(np.asarray(dmasks))
return dimage, dmasks
def outline_mask(mask,line_width=2):
outlines = np.zeros_like(mask[0])
for i in mask:
dill = morphology.binary_dilation(i,morphology.disk(line_width))
outlines[dill==1] = 1
#mind = morphology.binary_erosion(i,morphology.disk(line_width))
outlines[i==1] = 2
return outlines
if __name__ == '__main__':
im_path = "/mnt/Local_SSD/stefan/cell_data_sets/LIVECell_dataset_2021/images/livecell_train_val_images/SHSY5Y/"
an_path = "/mnt/Local_SSD/stefan/cell_data_sets/LIVECell_dataset_2021/annotations/LIVECell_single_cells/shsy5y/train.json"
out_path = "/mnt/Local_SSD/stefan/cell_data_sets/training/221219_livecell_1/pngs/"
image_id = 100
lims = (np.arange(1,10)*20).astype(int)
im_list = sorted(glob(im_path+"*.tif"))
image, masks = data_from_coco(im_path, an_path, idn=image_id)
dimages, dmasks = [],[]
for i in range(10):
di,dm = distort_all(image,masks,amp=10,radius=2.,shift=20,strength=0.1*i)
dimages.append(di)
dmasks.append(dm)
dimages, dmasks = np.stack(dimages), np.stack(dmasks)
dimages_left = dimages[:,:, :dimages.shape[1] ]
dimages_right = dimages[:,:, -dimages.shape[1]:]
dmasks_left = dmasks[:,:,:, :dimages.shape[1] ]
dmasks_right = dmasks[:,:,:, -dimages.shape[1]:]
image_crops = np.vstack([dimages_left,dimages_right])
mask_crops = np.vstack([dmasks_left ,dmasks_right ])
def process_all(ind):
size = dimages.shape[1]
lw = 2
ic = image_crops[ind].astype("uint8")
im = (outline_mask(mask_crops[ind],lw)*100).astype("uint8")
io.imsave(out_path+str(10000000+ind*100)[1:]+"raw.png" ,ic)
io.imsave(out_path+str(10000000+ind*100)[1:]+"mask.png" ,im)
for n,i in enumerate(lims):
ic = image_crops[ind]
ic = image_crop_resize(ic,c=i,size=size)
ic = ic.astype("uint8")
im = mask_crops[ind]
im = mask_crop_resize(im,c=i,size=size)
im = outline_mask(im,lw*size/(size-i))*100
im = im.astype(int)
io.imsave(out_path+str(10000000+ind*100+n)[1:]+"raw.png" ,ic)
io.imsave(out_path+str(10000000+ind*100+n)[1:]+"mask.png" ,im)
for j in list(range(1,4)):
io.imsave(out_path+str(10000000+ind*100+n+10*j)[1:]+"raw.png" ,np.rot90(ic,j))
io.imsave(out_path+str(10000000+ind*100+n+10*j)[1:]+"mask.png" ,np.rot90(im,j))
io.imsave(out_path+str(10000000+ind*100+n+10*j+30)[1:]+"raw.png" ,np.rot90(ic.T,j))
io.imsave(out_path+str(10000000+ind*100+n+10*j+30)[1:]+"mask.png" ,np.rot90(im.T,j))
io.imsave(out_path+str(10000000+ind*100+n+10*j+60)[1:]+"raw.png" ,np.rot90(ic[::-1].T,j))
io.imsave(out_path+str(10000000+ind*100+n+10*j+60)[1:]+"mask.png" ,np.rot90(im[::-1].T,j))
_ = mCPU(process_all,range(len(image_crops)),40)