質問編集履歴
2
情報の追記
test
CHANGED
File without changes
|
test
CHANGED
@@ -154,6 +154,408 @@
|
|
154
154
|
|
155
155
|
|
156
156
|
|
157
|
+
以下は"ssd_mobilenet_v1_pets.config"の内容です.
|
158
|
+
|
159
|
+
```ssd_mobilenet_v1_pets.config
|
160
|
+
|
161
|
+
# SSD with Mobilenet v1, configured for Oxford-IIIT Pets Dataset.
|
162
|
+
|
163
|
+
# Users should configure the fine_tune_checkpoint field in the train config as
|
164
|
+
|
165
|
+
# well as the label_map_path and input_path fields in the train_input_reader and
|
166
|
+
|
167
|
+
# eval_input_reader. Search for "PATH_TO_BE_CONFIGURED" to find the fields that
|
168
|
+
|
169
|
+
# should be configured.
|
170
|
+
|
171
|
+
|
172
|
+
|
173
|
+
model {
|
174
|
+
|
175
|
+
ssd {
|
176
|
+
|
177
|
+
num_classes: 1
|
178
|
+
|
179
|
+
box_coder {
|
180
|
+
|
181
|
+
faster_rcnn_box_coder {
|
182
|
+
|
183
|
+
y_scale: 10.0
|
184
|
+
|
185
|
+
x_scale: 10.0
|
186
|
+
|
187
|
+
height_scale: 5.0
|
188
|
+
|
189
|
+
width_scale: 5.0
|
190
|
+
|
191
|
+
}
|
192
|
+
|
193
|
+
}
|
194
|
+
|
195
|
+
matcher {
|
196
|
+
|
197
|
+
argmax_matcher {
|
198
|
+
|
199
|
+
matched_threshold: 0.5
|
200
|
+
|
201
|
+
unmatched_threshold: 0.5
|
202
|
+
|
203
|
+
ignore_thresholds: false
|
204
|
+
|
205
|
+
negatives_lower_than_unmatched: true
|
206
|
+
|
207
|
+
force_match_for_each_row: true
|
208
|
+
|
209
|
+
}
|
210
|
+
|
211
|
+
}
|
212
|
+
|
213
|
+
similarity_calculator {
|
214
|
+
|
215
|
+
iou_similarity {
|
216
|
+
|
217
|
+
}
|
218
|
+
|
219
|
+
}
|
220
|
+
|
221
|
+
anchor_generator {
|
222
|
+
|
223
|
+
ssd_anchor_generator {
|
224
|
+
|
225
|
+
num_layers: 6
|
226
|
+
|
227
|
+
min_scale: 0.2
|
228
|
+
|
229
|
+
max_scale: 0.95
|
230
|
+
|
231
|
+
aspect_ratios: 1.0
|
232
|
+
|
233
|
+
aspect_ratios: 2.0
|
234
|
+
|
235
|
+
aspect_ratios: 0.5
|
236
|
+
|
237
|
+
aspect_ratios: 3.0
|
238
|
+
|
239
|
+
aspect_ratios: 0.3333
|
240
|
+
|
241
|
+
}
|
242
|
+
|
243
|
+
}
|
244
|
+
|
245
|
+
image_resizer {
|
246
|
+
|
247
|
+
fixed_shape_resizer {
|
248
|
+
|
249
|
+
height: 300
|
250
|
+
|
251
|
+
width: 300
|
252
|
+
|
253
|
+
}
|
254
|
+
|
255
|
+
}
|
256
|
+
|
257
|
+
box_predictor {
|
258
|
+
|
259
|
+
convolutional_box_predictor {
|
260
|
+
|
261
|
+
min_depth: 0
|
262
|
+
|
263
|
+
max_depth: 0
|
264
|
+
|
265
|
+
num_layers_before_predictor: 0
|
266
|
+
|
267
|
+
use_dropout: false
|
268
|
+
|
269
|
+
dropout_keep_probability: 0.8
|
270
|
+
|
271
|
+
kernel_size: 1
|
272
|
+
|
273
|
+
box_code_size: 4
|
274
|
+
|
275
|
+
apply_sigmoid_to_scores: false
|
276
|
+
|
277
|
+
conv_hyperparams {
|
278
|
+
|
279
|
+
activation: RELU_6,
|
280
|
+
|
281
|
+
regularizer {
|
282
|
+
|
283
|
+
l2_regularizer {
|
284
|
+
|
285
|
+
weight: 0.00004
|
286
|
+
|
287
|
+
}
|
288
|
+
|
289
|
+
}
|
290
|
+
|
291
|
+
initializer {
|
292
|
+
|
293
|
+
truncated_normal_initializer {
|
294
|
+
|
295
|
+
stddev: 0.03
|
296
|
+
|
297
|
+
mean: 0.0
|
298
|
+
|
299
|
+
}
|
300
|
+
|
301
|
+
}
|
302
|
+
|
303
|
+
batch_norm {
|
304
|
+
|
305
|
+
train: true,
|
306
|
+
|
307
|
+
scale: true,
|
308
|
+
|
309
|
+
center: true,
|
310
|
+
|
311
|
+
decay: 0.9997,
|
312
|
+
|
313
|
+
epsilon: 0.001,
|
314
|
+
|
315
|
+
}
|
316
|
+
|
317
|
+
}
|
318
|
+
|
319
|
+
}
|
320
|
+
|
321
|
+
}
|
322
|
+
|
323
|
+
feature_extractor {
|
324
|
+
|
325
|
+
type: 'ssd_mobilenet_v1'
|
326
|
+
|
327
|
+
min_depth: 16
|
328
|
+
|
329
|
+
depth_multiplier: 1.0
|
330
|
+
|
331
|
+
conv_hyperparams {
|
332
|
+
|
333
|
+
activation: RELU_6,
|
334
|
+
|
335
|
+
regularizer {
|
336
|
+
|
337
|
+
l2_regularizer {
|
338
|
+
|
339
|
+
weight: 0.00004
|
340
|
+
|
341
|
+
}
|
342
|
+
|
343
|
+
}
|
344
|
+
|
345
|
+
initializer {
|
346
|
+
|
347
|
+
truncated_normal_initializer {
|
348
|
+
|
349
|
+
stddev: 0.03
|
350
|
+
|
351
|
+
mean: 0.0
|
352
|
+
|
353
|
+
}
|
354
|
+
|
355
|
+
}
|
356
|
+
|
357
|
+
batch_norm {
|
358
|
+
|
359
|
+
train: true,
|
360
|
+
|
361
|
+
scale: true,
|
362
|
+
|
363
|
+
center: true,
|
364
|
+
|
365
|
+
decay: 0.9997,
|
366
|
+
|
367
|
+
epsilon: 0.001,
|
368
|
+
|
369
|
+
}
|
370
|
+
|
371
|
+
}
|
372
|
+
|
373
|
+
}
|
374
|
+
|
375
|
+
loss {
|
376
|
+
|
377
|
+
classification_loss {
|
378
|
+
|
379
|
+
weighted_sigmoid {
|
380
|
+
|
381
|
+
anchorwise_output: true
|
382
|
+
|
383
|
+
}
|
384
|
+
|
385
|
+
}
|
386
|
+
|
387
|
+
localization_loss {
|
388
|
+
|
389
|
+
weighted_smooth_l1 {
|
390
|
+
|
391
|
+
anchorwise_output: true
|
392
|
+
|
393
|
+
}
|
394
|
+
|
395
|
+
}
|
396
|
+
|
397
|
+
hard_example_miner {
|
398
|
+
|
399
|
+
num_hard_examples: 3000
|
400
|
+
|
401
|
+
iou_threshold: 0.99
|
402
|
+
|
403
|
+
loss_type: CLASSIFICATION
|
404
|
+
|
405
|
+
max_negatives_per_positive: 3
|
406
|
+
|
407
|
+
min_negatives_per_image: 0
|
408
|
+
|
409
|
+
}
|
410
|
+
|
411
|
+
classification_weight: 1.0
|
412
|
+
|
413
|
+
localization_weight: 1.0
|
414
|
+
|
415
|
+
}
|
416
|
+
|
417
|
+
normalize_loss_by_num_matches: true
|
418
|
+
|
419
|
+
post_processing {
|
420
|
+
|
421
|
+
batch_non_max_suppression {
|
422
|
+
|
423
|
+
score_threshold: 1e-8
|
424
|
+
|
425
|
+
iou_threshold: 0.6
|
426
|
+
|
427
|
+
max_detections_per_class: 100
|
428
|
+
|
429
|
+
max_total_detections: 100
|
430
|
+
|
431
|
+
}
|
432
|
+
|
433
|
+
score_converter: SIGMOID
|
434
|
+
|
435
|
+
}
|
436
|
+
|
437
|
+
}
|
438
|
+
|
439
|
+
}
|
440
|
+
|
441
|
+
|
442
|
+
|
443
|
+
train_config: {
|
444
|
+
|
445
|
+
batch_size: 24
|
446
|
+
|
447
|
+
optimizer {
|
448
|
+
|
449
|
+
rms_prop_optimizer: {
|
450
|
+
|
451
|
+
learning_rate: {
|
452
|
+
|
453
|
+
exponential_decay_learning_rate {
|
454
|
+
|
455
|
+
initial_learning_rate: 0.004
|
456
|
+
|
457
|
+
decay_steps: 800720
|
458
|
+
|
459
|
+
decay_factor: 0.95
|
460
|
+
|
461
|
+
}
|
462
|
+
|
463
|
+
}
|
464
|
+
|
465
|
+
momentum_optimizer_value: 0.9
|
466
|
+
|
467
|
+
decay: 0.9
|
468
|
+
|
469
|
+
epsilon: 1.0
|
470
|
+
|
471
|
+
}
|
472
|
+
|
473
|
+
}
|
474
|
+
|
475
|
+
fine_tune_checkpoint: "ssd_mobilenet_v1_coco_11_06_2017/model.ckpt"
|
476
|
+
|
477
|
+
from_detection_checkpoint: true
|
478
|
+
|
479
|
+
# Note: The below line limits the training process to 200K steps, which we
|
480
|
+
|
481
|
+
# empirically found to be sufficient enough to train the pets dataset. This
|
482
|
+
|
483
|
+
# effectively bypasses the learning rate schedule (the learning rate will
|
484
|
+
|
485
|
+
# never decay). Remove the below line to train indefinitely.
|
486
|
+
|
487
|
+
num_steps: 200000
|
488
|
+
|
489
|
+
data_augmentation_options {
|
490
|
+
|
491
|
+
random_horizontal_flip {
|
492
|
+
|
493
|
+
}
|
494
|
+
|
495
|
+
}
|
496
|
+
|
497
|
+
data_augmentation_options {
|
498
|
+
|
499
|
+
ssd_random_crop {
|
500
|
+
|
501
|
+
}
|
502
|
+
|
503
|
+
}
|
504
|
+
|
505
|
+
}
|
506
|
+
|
507
|
+
|
508
|
+
|
509
|
+
train_input_reader: {
|
510
|
+
|
511
|
+
tf_record_input_reader {
|
512
|
+
|
513
|
+
input_path: "data/train.record"
|
514
|
+
|
515
|
+
}
|
516
|
+
|
517
|
+
label_map_path: "data/object-detection.pbtxt"
|
518
|
+
|
519
|
+
}
|
520
|
+
|
521
|
+
|
522
|
+
|
523
|
+
eval_config: {
|
524
|
+
|
525
|
+
num_examples: 2000
|
526
|
+
|
527
|
+
# Note: The below line limits the evaluation process to 10 evaluations.
|
528
|
+
|
529
|
+
# Remove the below line to evaluate indefinitely.
|
530
|
+
|
531
|
+
max_evals: 10
|
532
|
+
|
533
|
+
}
|
534
|
+
|
535
|
+
|
536
|
+
|
537
|
+
eval_input_reader: {
|
538
|
+
|
539
|
+
tf_record_input_reader {
|
540
|
+
|
541
|
+
input_path: "data/test.record"
|
542
|
+
|
543
|
+
}
|
544
|
+
|
545
|
+
label_map_path: "data/object-detection.pbtxt"
|
546
|
+
|
547
|
+
shuffle: false
|
548
|
+
|
549
|
+
num_readers: 1
|
550
|
+
|
551
|
+
}
|
552
|
+
|
553
|
+
|
554
|
+
|
555
|
+
```
|
556
|
+
|
557
|
+
|
558
|
+
|
157
559
|
|
158
560
|
|
159
561
|
|
1
説明の追加
test
CHANGED
File without changes
|
test
CHANGED
@@ -26,21 +26,21 @@
|
|
26
26
|
|
27
27
|
-images
|
28
28
|
|
29
|
-
--image
|
29
|
+
--images.png
|
30
30
|
|
31
|
-
--image
|
31
|
+
--images.xml
|
32
32
|
|
33
33
|
--test
|
34
34
|
|
35
|
-
---image
|
35
|
+
---images.png
|
36
36
|
|
37
|
-
---image
|
37
|
+
---images.xml
|
38
38
|
|
39
39
|
--train
|
40
40
|
|
41
|
-
---image
|
41
|
+
---images.png
|
42
42
|
|
43
|
-
---image
|
43
|
+
---images.xml
|
44
44
|
|
45
45
|
-data
|
46
46
|
|
@@ -70,7 +70,7 @@
|
|
70
70
|
|
71
71
|
上記のものを移動しました.
|
72
72
|
|
73
|
-
|
73
|
+
imagesの中に入っているimages.png は複数の画像が入っています.
|
74
74
|
|
75
75
|
```
|
76
76
|
|
@@ -144,6 +144,18 @@
|
|
144
144
|
|
145
145
|
メッセージからも正しく学習が開始されているように見えませんでした.
|
146
146
|
|
147
|
+
batch_size:24
|
148
|
+
|
149
|
+
num_terms:20000
|
150
|
+
|
151
|
+
|
152
|
+
|
153
|
+
"INFO:tensorflow:global_step/sec: 0" が表示されるまでにおよそ1分ほどかかっています.
|
154
|
+
|
155
|
+
|
156
|
+
|
157
|
+
|
158
|
+
|
147
159
|
|
148
160
|
|
149
161
|
環境
|