-
Notifications
You must be signed in to change notification settings - Fork 79
/
utils.py
389 lines (331 loc) · 17.8 KB
/
utils.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
# written by Xiaohui Zhao
# 2018-01
import numpy as np
import csv
from os.path import join
try:
import cv2
except ImportError:
pass
c_threshold = 0.5
def cal_accuracy(data_loader, grid_table, gt_classes, model_output_val, label_mapids, bbox_mapids):
#num_tp = 0
#num_fn = 0
res = ''
num_correct = 0
num_correct_strict = 0
num_correct_soft = 0
num_all = grid_table.shape[0] * (model_output_val.shape[-1]-1)
for b in range(grid_table.shape[0]):
data_input_flat = grid_table[b,:,:,0].reshape([-1])
labels = gt_classes[b,:,:].reshape([-1])
logits = model_output_val[b,:,:,:].reshape([-1, data_loader.num_classes])
label_mapid = label_mapids[b]
bbox_mapid = bbox_mapids[b]
rows, cols = grid_table.shape[1:3]
bbox_id = np.array([row*cols+col for row in range(rows) for col in range(cols)])
# ignore inputs that are not word
indexes = np.where(data_input_flat != 0)[0]
data_selected = data_input_flat[indexes]
labels_selected = labels[indexes]
logits_array_selected = logits[indexes]
bbox_id_selected = bbox_id[indexes]
# calculate accuracy
#test_classes = [1,2,3,4,5]
#for c in test_classes:
for c in range(1, data_loader.num_classes):
labels_indexes = np.where(labels_selected == c)[0]
logits_indexes = np.where(logits_array_selected[:,c] > c_threshold)[0]
labels_words = list(data_loader.index_to_word[i] for i in data_selected[labels_indexes])
logits_words = list(data_loader.index_to_word[i] for i in data_selected[logits_indexes])
label_bbox_ids = label_mapid[c] # GT bbox_ids related to the type of class
logit_bbox_ids = [bbox_mapid[bbox] for bbox in bbox_id_selected[logits_indexes] if bbox in bbox_mapid]
#if np.array_equal(labels_indexes, logits_indexes):
if set(label_bbox_ids) == set(logit_bbox_ids): # decide as correct when all ids match
num_correct_strict += 1
num_correct_soft += 1
elif set(label_bbox_ids).issubset(set(logit_bbox_ids)): # correct when gt is subset of gt
num_correct_soft += 1
try: # calculate prevalence with decimal precision
num_correct += np.shape(np.intersect1d(labels_indexes, logits_indexes))[0] / np.shape(labels_indexes)[0]
except ZeroDivisionError:
if np.shape(labels_indexes)[0] == 0:
num_correct += 1
else:
num_correct += 0
# show results without the <DontCare> class
if b==0:
res += '\n{}(GT/Inf):\t"'.format(data_loader.classes[c])
# ground truth label
res += ' '.join(data_loader.index_to_word[i] for i in data_selected[labels_indexes])
res += '" | "'
res += ' '.join(data_loader.index_to_word[i] for i in data_selected[logits_indexes])
res += '"'
# wrong inferences results
if not np.array_equal(labels_indexes, logits_indexes):
res += '\n \t FALSES =>>'
logits_flat = logits_array_selected[:,c]
fault_logits_indexes = np.setdiff1d(logits_indexes, labels_indexes)
for i in range(len(data_selected)):
if i not in fault_logits_indexes: # only show fault_logits_indexes
continue
w = data_loader.index_to_word[data_selected[i]]
l = data_loader.classes[labels_selected[i]]
res += ' "%s"/%s, '%(w, l)
#res += ' "%s"/%.2f%s, '%(w, logits_flat[i], l)
#print(res)
prevalence = num_correct / num_all
accuracy_strict = num_correct_strict / num_all
accuracy_soft = num_correct_soft / num_all
return prevalence, accuracy_strict, accuracy_soft, res.encode("utf-8")
def cal_save_results(data_loader, grid_table, gt_classes, model_output_val, label_mapids, bbox_mapids, file_names, save_prefix):
res = ''
num_correct = 0
num_correct_strict = 0
num_correct_soft = 0
num_all = grid_table.shape[0] * (model_output_val.shape[-1]-1)
for b in range(grid_table.shape[0]):
filename = file_names[0]
data_input_flat = grid_table[b,:,:,0].reshape([-1])
labels = gt_classes[b,:,:].reshape([-1])
logits = model_output_val[b,:,:,:].reshape([-1, data_loader.num_classes])
label_mapid = label_mapids[b]
bbox_mapid = bbox_mapids[b]
rows, cols = grid_table.shape[1:3]
bbox_id = np.array([row*cols+col for row in range(rows) for col in range(cols)])
# ignore inputs that are not word
indexes = np.where(data_input_flat != 0)[0]
data_selected = data_input_flat[indexes]
labels_selected = labels[indexes]
logits_array_selected = logits[indexes]
bbox_id_selected = bbox_id[indexes]
# calculate accuracy
for c in range(1, data_loader.num_classes):
labels_indexes = np.where(labels_selected == c)[0]
logits_indexes = np.where(logits_array_selected[:,c] > c_threshold)[0]
labels_words = list(data_loader.index_to_word[i] for i in data_selected[labels_indexes])
logits_words = list(data_loader.index_to_word[i] for i in data_selected[logits_indexes])
label_bbox_ids = label_mapid[c] # GT bbox_ids related to the type of class
logit_bbox_ids = [bbox_mapid[bbox] for bbox in bbox_id_selected[logits_indexes] if bbox in bbox_mapid]
#if np.array_equal(labels_indexes, logits_indexes):
if set(label_bbox_ids) == set(logit_bbox_ids): # decide as correct when all ids match
num_correct_strict += 1
num_correct_soft += 1
elif set(label_bbox_ids).issubset(set(logit_bbox_ids)): # correct when gt is subset of gt
num_correct_soft += 1
try: # calculate prevalence with decimal precision
num_correct += np.shape(np.intersect1d(labels_indexes, logits_indexes))[0] / np.shape(labels_indexes)[0]
except ZeroDivisionError:
if np.shape(labels_indexes)[0] == 0:
num_correct += 1
else:
num_correct += 0
# show results without the <DontCare> class
# ground truth label
gt = str(' '.join(data_loader.index_to_word[i] for i in data_selected[labels_indexes]))
predict = str(' '.join(data_loader.index_to_word[i] for i in data_selected[logits_indexes]))
# write results to csv
fieldnames = ['TaskID', 'GT', 'Predicted']
csv_filename = 'data/results/' + save_prefix + '_' + data_loader.classes[c] + '.csv'
writer = csv.DictWriter(open(csv_filename, 'a'), fieldnames=fieldnames)
row = {'TaskID':filename, 'GT':gt, 'Predicted':predict}
writer.writerow(row)
csv_diff_filename = 'data/results/' + save_prefix + '_Diff_' + data_loader.classes[c] + '.csv'
if gt != predict:
writer = csv.DictWriter(open(csv_diff_filename, 'a'), fieldnames=fieldnames)
row = {'TaskID':filename, 'GT':gt, 'Predicted':predict}
writer.writerow(row)
if b == 0:
res += '\n{}(GT/Inf):\t"'.format(data_loader.classes[c])
res += gt + '" | "' + predict + '"'
# wrong inferences results
if not np.array_equal(labels_indexes, logits_indexes):
res += '\n \t FALSES =>>'
logits_flat = logits_array_selected[:,c]
fault_logits_indexes = np.setdiff1d(logits_indexes, labels_indexes)
for i in range(len(data_selected)):
if i not in fault_logits_indexes: # only show fault_logits_indexes
continue
w = data_loader.index_to_word[data_selected[i]]
l = data_loader.classes[labels_selected[i]]
res += ' "%s"/%s, '%(w, l)
#res += ' "%s"/%.2f%s, '%(w, logits_flat[i], l)
#print(res)
prevalence = num_correct / num_all
accuracy_strict = num_correct_strict / num_all
accuracy_soft = num_correct_soft / num_all
return prevalence, accuracy_strict, accuracy_soft, res.encode("utf-8")
def vis_bbox(data_loader, file_prefix, grid_table, gt_classes, model_output_val, file_name, bboxes, shape):
data_input_flat = grid_table.reshape([-1])
labels = gt_classes.reshape([-1])
logits = model_output_val.reshape([-1, data_loader.num_classes])
bboxes = bboxes.reshape([-1])
max_len = 768*2 # upper boundary of image display size
img = cv2.imread(join(file_prefix, file_name))
if img is not None:
shape = list(img.shape)
bbox_pad = 1
gt_color = [[255, 250, 240], [152, 245, 255], [119,204,119], [100, 149, 237],
[192, 255, 62], [119,119,204], [114,124,114], [240, 128, 128], [255, 105, 180]]
inf_color = [[255, 222, 173], [0, 255, 255], [50,219,50], [72, 61, 139],
[154, 205, 50], [50,50,219], [64,76,64], [255, 0, 0], [255, 20, 147]]
font_size = 0.5
font = cv2.FONT_HERSHEY_COMPLEX
ft_color = [50, 50, 250]
factor = max_len / max(shape)
shape[0], shape[1] = [int(s*factor) for s in shape[:2]]
img = cv2.resize(img, (shape[1], shape[0]))
overlay_box = np.zeros(shape, dtype=img.dtype)
overlay_line = np.zeros(shape, dtype=img.dtype)
for i in range(len(data_input_flat)):
if len(bboxes[i]) > 0:
x,y,w,h = [int(p*factor) for p in bboxes[i]]
else:
row = i // data_loader.rows
col = i % data_loader.cols
x = shape[1] // data_loader.cols * col
y = shape[0] // data_loader.rows * row
w = shape[1] // data_loader.cols * 2
h = shape[0] // data_loader.cols * 2
if data_input_flat[i] and labels[i]:
gt_id = labels[i]
cv2.rectangle(overlay_box, (x,y), (x+w,y+h), gt_color[gt_id], -1)
if max(logits[i]) > c_threshold:
inf_id = np.argmax(logits[i])
if inf_id:
cv2.rectangle(overlay_line, (x+bbox_pad,y+bbox_pad), \
(x+bbox_pad+w,y+bbox_pad+h), inf_color[inf_id], max_len//768*2)
#text = data_loader.classes[gt_id] + '|' + data_loader.classes[inf_id]
#cv2.putText(img, text, (x,y), font, font_size, ft_color)
# legends
w = shape[1] // data_loader.cols * 4
h = shape[0] // data_loader.cols * 2
for i in range(1, len(data_loader.classes)):
row = i * 3
col = 0
x = shape[1] // data_loader.cols * col
y = shape[0] // data_loader.rows * row
cv2.rectangle(img, (x,y), (x+w,y+h), gt_color[i], -1)
cv2.putText(img, data_loader.classes[i], (x+w,y+h), font, 0.8, ft_color)
row = i * 3 + 1
col = 0
x = shape[1] // data_loader.cols * col
y = shape[0] // data_loader.rows * row
cv2.rectangle(img, (x+bbox_pad,y+bbox_pad), \
(x+bbox_pad+w,y+bbox_pad+h), inf_color[i], max_len//384)
alpha = 0.4
cv2.addWeighted(overlay_box, alpha, img, 1-alpha, 0, img)
cv2.addWeighted(overlay_line, 1-alpha, img, 1, 0, img)
cv2.imwrite('results/' + file_name[:-4]+'.png', img)
cv2.imshow("test", img)
cv2.waitKey(0)
def cal_accuracy_table(data_loader, grid_table, gt_classes, model_output_val, label_mapids, bbox_mapids):
#num_tp = 0
#num_fn = 0
res = ''
num_correct = 0
num_correct_strict = 0
num_correct_soft = 0
num_all = grid_table.shape[0] * (model_output_val.shape[-1]-1)
for b in range(grid_table.shape[0]):
data_input_flat = grid_table[b,:,:,0]
rows, cols = grid_table.shape[1:3]
labels = gt_classes[b,:,:]
logits = model_output_val[b,:,:,:].reshape([rows, cols, data_loader.num_classes])
label_mapid = label_mapids[b]
bbox_mapid = bbox_mapids[b]
bbox_id = np.array([row*cols+col for row in range(rows) for col in range(cols)])
# calculate accuracy
#test_classes = [1,2,3,4,5]
#for c in test_classes:
for c in range(1, data_loader.num_classes):
label_rows, label_cols = np.where(labels == c)
logit_rows, logit_cols = np.where(logits[:,:,c] > c_threshold)
if min(label_rows) == min(logit_rows) and max(label_cols) == max(logit_cols):
num_correct_strict += 1
num_correct_soft += 1
num_correct += 1
if min(label_rows) > min(logit_rows) and max(label_cols) < max(logit_cols):
num_correct_soft += 1
num_correct += 1
prevalence = num_correct / num_all
accuracy_strict = num_correct_strict / num_all
accuracy_soft = num_correct_soft / num_all
return prevalence, accuracy_strict, accuracy_soft, res.encode("utf-8")
def vis_table(data_loader, file_prefix, grid_table, gt_classes, model_output_val, file_name, bboxes, shape):
data_input_flat = grid_table.reshape([-1])
labels = gt_classes.reshape([-1])
logits = model_output_val.reshape([-1, data_loader.num_classes])
bboxes = bboxes.reshape([-1])
max_len = 768*2 # upper boundary of image display size
img = cv2.imread(join(file_prefix, file_name))
if img is not None:
shape = list(img.shape)
bbox_pad = 1
gt_color = [[255, 250, 240], [152, 245, 255], [119,204,119], [100, 149, 237],
[192, 255, 62], [119,119,204], [114,124,114], [240, 128, 128], [255, 105, 180]]
inf_color = [[255, 222, 173], [0, 255, 255], [50,219,50], [72, 61, 139],
[154, 205, 50], [50,50,219], [64,76,64], [255, 0, 0], [255, 20, 147]]
font_size = 0.5
font = cv2.FONT_HERSHEY_COMPLEX
ft_color = [50, 50, 250]
factor = max_len / max(shape)
shape[0], shape[1] = [int(s*factor) for s in shape[:2]]
img = cv2.resize(img, (shape[1], shape[0]))
overlay_box = np.zeros(shape, dtype=img.dtype)
overlay_line = np.zeros(shape, dtype=img.dtype)
gt_x, gt_y, gt_r, gt_b = 99999, 99999, 0, 0
inf_x, inf_y, inf_r, inf_b = 99999, 99999, 0, 0
for i in range(len(data_input_flat)):
if len(bboxes[i]) > 0:
x,y,w,h = [int(p*factor) for p in bboxes[i]]
else:
row = i // data_loader.rows
col = i % data_loader.cols
x = shape[1] // data_loader.cols * col
y = shape[0] // data_loader.rows * row
w = shape[1] // data_loader.cols * 2
h = shape[0] // data_loader.cols * 2
if data_input_flat[i] and labels[i]:
gt_id = labels[i]
cv2.rectangle(overlay_box, (x,y), (x+w,y+h), gt_color[gt_id], -1)
gt_x = min(x, gt_x)
gt_y = min(y, gt_y)
gt_r = max(x+w, gt_r)
gt_b = max(y+h, gt_b)
if max(logits[i]) > c_threshold:
inf_id = np.argmax(logits[i])
if inf_id:
cv2.rectangle(overlay_line, (x+bbox_pad,y+bbox_pad), \
(x+bbox_pad+w,y+bbox_pad+h), inf_color[inf_id], max_len//768*2)
inf_x = min(x, inf_x)
inf_y = min(y, inf_y)
inf_r = max(x+w, inf_r)
inf_b = max(y+h, inf_b)
#text = data_loader.classes[gt_id] + '|' + data_loader.classes[inf_id]
#cv2.putText(img, text, (x,y), font, font_size, ft_color)
cv2.rectangle(overlay_box, (gt_x,gt_y), (gt_r,gt_b), [180,180,215], -1)
cv2.rectangle(overlay_line, (inf_x+bbox_pad,inf_y+bbox_pad), (inf_r+bbox_pad,inf_b+bbox_pad), [0,115,255], max_len//768*2)
# legends
w = shape[1] // data_loader.cols * 4
h = shape[0] // data_loader.cols * 2
for i in range(1, len(data_loader.classes)):
row = i * 3
col = 0
x = shape[1] // data_loader.cols * col
y = shape[0] // data_loader.rows * row
cv2.rectangle(img, (x,y), (x+w,y+h), gt_color[i], -1)
cv2.putText(img, data_loader.classes[i], (x+w,y+h), font, 0.8, ft_color)
row = i * 3 + 1
col = 0
x = shape[1] // data_loader.cols * col
y = shape[0] // data_loader.rows * row
cv2.rectangle(img, (x+bbox_pad,y+bbox_pad), \
(x+bbox_pad+w,y+bbox_pad+h), inf_color[i], max_len//384)
alpha = 0.4
cv2.addWeighted(overlay_box, alpha, img, 1-alpha, 0, img)
cv2.addWeighted(overlay_line, 1-alpha, img, 1, 0, img)
cv2.imwrite('results/' + file_name[:-4]+'.png', img)
cv2.imshow("test", img)
cv2.waitKey(0)