-
Notifications
You must be signed in to change notification settings - Fork 28
/
fsaf_layers.py
406 lines (351 loc) · 17.3 KB
/
fsaf_layers.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
402
403
404
405
406
from keras.layers import Layer
import tensorflow as tf
from util_graphs import trim_zeros_graph, prop_box_graph, prop_box_graph_2
import keras.backend as K
from losses import focal, iou
from configure import MAX_NUM_GT_BOXES, STRIDES, POS_SCALE, IGNORE_SCALE
def level_select(cls_pred, regr_pred, gt_boxes, feature_shapes, strides, pos_scale=0.2):
"""
Args:
cls_pred: (sum(fh * fw), num_classes)
regr_pred: (sum(fh * fw), 4)
gt_boxes: (MAX_NUM_GT_BOXES, 5)
feature_shapes: (5, 2)
strides:
pos_scale:
Returns:
"""
gt_labels = tf.cast(gt_boxes[:, 4], tf.int32)
gt_boxes = gt_boxes[:, :4]
focal_loss = focal()
iou_loss = iou()
gt_boxes, non_zeros = trim_zeros_graph(gt_boxes)
num_gt_boxes = tf.shape(gt_boxes)[0]
gt_labels = tf.boolean_mask(gt_labels, non_zeros)
level_losses = []
for level_id in range(len(strides)):
stride = strides[level_id]
fh = feature_shapes[level_id][0]
fw = feature_shapes[level_id][1]
fa = tf.reduce_prod(feature_shapes, axis=-1)
start_idx = tf.reduce_sum(fa[:level_id])
end_idx = start_idx + fh * fw
cls_pred_i = tf.reshape(cls_pred[start_idx:end_idx], (fh, fw, tf.shape(cls_pred)[-1]))
regr_pred_i = tf.reshape(regr_pred[start_idx:end_idx], (fh, fw, tf.shape(regr_pred)[-1]))
proj_boxes = gt_boxes / stride
x1, y1, x2, y2 = prop_box_graph(proj_boxes, pos_scale, fw, fh)
def compute_gt_box_loss(args):
x1_ = args[0]
y1_ = args[1]
x2_ = args[2]
y2_ = args[3]
gt_box = args[4]
gt_label = args[5]
locs_cls_pred_i = cls_pred_i[y1_:y2_, x1_:x2_, :]
locs_cls_pred_i = tf.reshape(locs_cls_pred_i, (-1, tf.shape(locs_cls_pred_i)[-1]))
locs_cls_true_i = tf.zeros_like(locs_cls_pred_i)
gt_label_col = tf.ones_like(locs_cls_true_i[:, 0:1])
locs_cls_true_i = tf.concat([locs_cls_true_i[:, :gt_label],
gt_label_col,
locs_cls_true_i[:, gt_label + 1:],
], axis=-1)
loss_cls = focal_loss(K.expand_dims(locs_cls_true_i, axis=0), K.expand_dims(locs_cls_pred_i, axis=0))
locs_regr_pred_i = regr_pred_i[y1_:y2_, x1_:x2_, :]
locs_regr_pred_i = tf.reshape(locs_regr_pred_i, (-1, tf.shape(locs_regr_pred_i)[-1]))
locs_x = K.arange(x1_, x2_, dtype=tf.float32)
locs_y = K.arange(y1_, y2_, dtype=tf.float32)
shift_x = (locs_x + 0.5) * stride
shift_y = (locs_y + 0.5) * stride
shift_xx, shift_yy = tf.meshgrid(shift_x, shift_y)
shift_xx = tf.reshape(shift_xx, (-1,))
shift_yy = tf.reshape(shift_yy, (-1,))
shifts = K.stack((shift_xx, shift_yy, shift_xx, shift_yy), axis=-1)
l = tf.maximum(shifts[:, 0] - gt_box[0], 0)
t = tf.maximum(shifts[:, 1] - gt_box[1], 0)
r = tf.maximum(gt_box[2] - shifts[:, 2], 0)
b = tf.maximum(gt_box[3] - shifts[:, 3], 0)
locs_regr_true_i = tf.stack([l, t, r, b], axis=-1)
locs_regr_true_i /= 4.0
loss_regr = iou_loss(K.expand_dims(locs_regr_true_i, axis=0), K.expand_dims(locs_regr_pred_i, axis=0))
return loss_cls + loss_regr
level_loss = tf.map_fn(
compute_gt_box_loss,
elems=[x1, y1, x2, y2, gt_boxes, gt_labels],
dtype=tf.float32
)
level_losses.append(level_loss)
losses = tf.stack(level_losses, axis=-1)
gt_box_levels = tf.argmin(losses, axis=-1)
padding_gt_box_levels = tf.ones((MAX_NUM_GT_BOXES - num_gt_boxes), dtype=tf.int64) * -1
gt_box_levels = tf.concat([gt_box_levels, padding_gt_box_levels], axis=0)
return gt_box_levels
class LevelSelect(Layer):
def __init__(self, **kwargs):
super(LevelSelect, self).__init__(**kwargs)
def call(self, inputs, **kwargs):
batch_cls_pred = inputs[0]
batch_regr_pred = inputs[1]
feature_shapes = inputs[2][0]
batch_gt_boxes = inputs[3]
def _level_select(args):
cls_pred = args[0]
regr_pred = args[1]
gt_boxes = args[2]
return level_select(
cls_pred,
regr_pred,
gt_boxes,
feature_shapes=feature_shapes,
strides=STRIDES,
pos_scale=POS_SCALE
)
outputs = tf.map_fn(
_level_select,
elems=[batch_cls_pred, batch_regr_pred, batch_gt_boxes],
dtype=tf.int64,
)
return outputs
def compute_output_shape(self, input_shape):
"""
Computes the output shapes given the input shapes.
Args
input_shape : List of shapes of [batch_cls_pred, batch_regr_pred, feature_shapes, batch_gt_boxes].
Returns
shape of batch_gt_box_levels
"""
# return input_shape[0][0], config.MAX_NUM_GT_BOXES
return input_shape[0][0], None
def get_config(self):
"""
Gets the configuration of this layer.
Returns
Dictionary containing the parameters of this layer.
"""
config = super(LevelSelect, self).get_config()
return config
def build_fsaf_target(gt_box_levels, gt_boxes, feature_shapes, num_classes, strides, pos_scale, ignore_scale):
gt_labels = tf.cast(gt_boxes[:, 4], tf.int32)
gt_boxes = gt_boxes[:, :4]
cls_target = tf.zeros((0, num_classes))
cls_mask = tf.zeros((0,), dtype=tf.bool)
cls_num_pos = tf.zeros((0,))
regr_target = tf.zeros((0, 4))
regr_mask = tf.zeros((0,), dtype=tf.bool)
for level_id in range(len(strides)):
feature_shape = feature_shapes[level_id]
stride = strides[level_id]
fh = feature_shape[0]
fw = feature_shape[1]
level_gt_box_indices = tf.where(tf.equal(gt_box_levels, level_id))
def do_level_has_gt_boxes():
level_gt_boxes = tf.gather(gt_boxes, level_gt_box_indices[:, 0])
level_proj_boxes = level_gt_boxes / stride
level_gt_labels = tf.gather_nd(gt_labels, level_gt_box_indices)
ign_x1, ign_y1, ign_x2, ign_y2 = prop_box_graph_2(level_proj_boxes, ignore_scale, fw, fh)
pos_x1, pos_y1, pos_x2, pos_y2 = prop_box_graph_2(level_proj_boxes, pos_scale, fw, fh)
def build_single_gt_box_fsaf_target(args):
ign_x1_ = args[0]
ign_y1_ = args[1]
ign_x2_ = args[2]
ign_y2_ = args[3]
pos_x1_ = args[4]
pos_y1_ = args[5]
pos_x2_ = args[6]
pos_y2_ = args[7]
gt_box = args[8]
gt_label = args[9]
level_box_cls_target = tf.zeros((pos_y2_[0] - pos_y1_[0], pos_x2_[0] - pos_x1_[0], num_classes))
level_box_gt_label_col = tf.ones((pos_y2_[0] - pos_y1_[0], pos_x2_[0] - pos_x1_[0], 1))
level_box_cls_target = tf.concat((level_box_cls_target[..., :gt_label],
level_box_gt_label_col,
level_box_cls_target[..., gt_label + 1:]), axis=-1)
level_box_cls_pos_mask = tf.ones((pos_y2_[0] - pos_y1_[0], pos_x2_[0] - pos_x1_[0])) * 2.
ign_top_bot = tf.concat((pos_y1_ - ign_y1_, ign_y2_ - pos_y2_), axis=0)
ign_lef_rit = tf.concat((pos_x1_ - ign_x1_, ign_x2_ - pos_x2_), axis=0)
ign_pad = tf.stack([ign_top_bot, ign_lef_rit], axis=0)
level_box_cls_ign_mask = tf.pad(level_box_cls_pos_mask, ign_pad)
other_top_bot = tf.concat((ign_y1_, fh - ign_y2_), axis=0)
other_lef_rit = tf.concat((ign_x1_, fw - ign_x2_), axis=0)
other_pad = tf.stack([other_top_bot, other_lef_rit], axis=0)
level_box_cls_mask = tf.pad(level_box_cls_ign_mask, other_pad, constant_values=-1.)
level_box_cls_target = tf.pad(level_box_cls_target,
tf.concat((ign_pad + other_pad, tf.constant([[0, 0]])), axis=0))
locs_x = K.arange(pos_x1_[0], pos_x2_[0], dtype=tf.float32)
locs_y = K.arange(pos_y1_[0], pos_y2_[0], dtype=tf.float32)
shift_x = (locs_x + 0.5) * stride
shift_y = (locs_y + 0.5) * stride
shift_xx, shift_yy = tf.meshgrid(shift_x, shift_y)
shifts = K.stack((shift_xx, shift_yy, shift_xx, shift_yy), axis=-1)
l = tf.maximum(shifts[:, :, 0] - gt_box[0], 0)
t = tf.maximum(shifts[:, :, 1] - gt_box[1], 0)
r = tf.maximum(gt_box[2] - shifts[:, :, 2], 0)
b = tf.maximum(gt_box[3] - shifts[:, :, 3], 0)
deltas = K.stack((l, t, r, b), axis=-1)
level_box_regr_pos_target = deltas / 4.0
level_box_regr_pos_mask = tf.ones((pos_y2_[0] - pos_y1_[0], pos_x2_[0] - pos_x1_[0]))
level_box_regr_mask = tf.pad(level_box_regr_pos_mask, ign_pad + other_pad)
level_box_regr_target = tf.pad(level_box_regr_pos_target,
tf.concat((ign_pad + other_pad, tf.constant([[0, 0]])), axis=0))
level_box_pos_area = (l + r) * (t + b)
level_box_area = tf.pad(level_box_pos_area, ign_pad + other_pad, constant_values=1e7)
return level_box_cls_target, level_box_cls_mask, level_box_regr_target, level_box_regr_mask, level_box_area
level_cls_target, level_cls_mask, level_regr_target, level_regr_mask, level_area = tf.map_fn(
build_single_gt_box_fsaf_target,
elems=[
ign_x1, ign_y1, ign_x2, ign_y2,
pos_x1, pos_y1, pos_x2, pos_y2,
level_gt_boxes, level_gt_labels],
dtype=(tf.float32, tf.float32, tf.float32, tf.float32, tf.float32)
)
level_min_area_box_indices = tf.argmin(level_area, axis=0, output_type=tf.int32)
level_min_area_box_indices = tf.reshape(level_min_area_box_indices, (-1,))
locs_x = K.arange(0, fw)
locs_y = K.arange(0, fh)
locs_xx, locs_yy = tf.meshgrid(locs_x, locs_y)
locs_xx = tf.reshape(locs_xx, (-1,))
locs_yy = tf.reshape(locs_yy, (-1,))
level_indices = tf.stack((level_min_area_box_indices, locs_yy, locs_xx), axis=-1)
level_cls_target_ = tf.gather_nd(level_cls_target, level_indices)
level_regr_target_ = tf.gather_nd(level_regr_target, level_indices)
level_cls_num_pos_ = tf.reduce_sum(tf.cast(tf.equal(tf.reduce_max(level_cls_mask, axis=0), 2), tf.float32))
level_cls_mask = tf.equal(tf.reduce_max(level_cls_mask, axis=0), 2) | tf.equal(
tf.reduce_max(level_cls_mask, axis=0),
-1)
level_cls_mask_ = tf.reshape(level_cls_mask, (fh * fw,))
level_regr_mask = tf.reduce_sum(level_regr_mask, axis=0) > 0
level_regr_mask_ = tf.reshape(level_regr_mask, (fh * fw,))
return level_cls_target_, level_cls_mask_, level_cls_num_pos_, level_regr_target_, level_regr_mask_
def do_level_has_no_gt_boxes():
level_cls_target_ = tf.zeros((fh * fw, num_classes))
level_cls_mask_ = tf.ones((fh * fw,), dtype=tf.bool)
level_cls_num_pos_ = tf.zeros(())
level_regr_target_ = tf.zeros((fh * fw, 4))
level_regr_mask_ = tf.zeros((fh * fw,), dtype=tf.bool)
return level_cls_target_, level_cls_mask_, level_cls_num_pos_, level_regr_target_, level_regr_mask_
level_cls_target, level_cls_mask, level_cls_num_pos, level_regr_target, level_regr_mask = tf.cond(
tf.equal(tf.size(level_gt_box_indices), 0),
do_level_has_no_gt_boxes,
do_level_has_gt_boxes)
cls_target = tf.concat([cls_target, level_cls_target], axis=0)
cls_mask = tf.concat([cls_mask, level_cls_mask], axis=0)
cls_num_pos = tf.concat([cls_num_pos, level_cls_num_pos[None]], axis=0)
regr_target = tf.concat([regr_target, level_regr_target], axis=0)
regr_mask = tf.concat([regr_mask, level_regr_mask], axis=0)
cls_num_pos = tf.reduce_sum(cls_num_pos)
return [cls_target, cls_mask, cls_num_pos, regr_target, regr_mask]
class FSAFTarget(Layer):
def __init__(self, num_classes, **kwargs):
super(FSAFTarget, self).__init__(**kwargs)
self.num_classes = num_classes
def call(self, inputs, **kwargs):
batch_gt_box_levels = inputs[0]
feature_shapes = inputs[1][0]
batch_gt_boxes = inputs[2]
def _build_fsaf_target(args):
gt_box_levels = args[0]
gt_boxes = args[1]
return build_fsaf_target(
gt_box_levels,
gt_boxes,
feature_shapes=feature_shapes,
num_classes=self.num_classes,
strides=STRIDES,
pos_scale=POS_SCALE,
ignore_scale=IGNORE_SCALE,
)
outputs = tf.map_fn(
_build_fsaf_target,
elems=[batch_gt_box_levels, batch_gt_boxes],
dtype=[tf.float32, tf.bool, tf.float32, tf.float32, tf.bool],
)
return outputs
def compute_output_shape(self, input_shape):
"""
Computes the output shapes given the input shapes.
Args
input_shape : List of shapes of [batch_gt_box_levels, feature_shapes, batch_gt_boxes].
Returns
List of tuples representing the shapes of [batch_cls_target, batch_cls_mask, batch_num_pos, batch_regr_target, batch_regr_mask]
"""
batch_size = input_shape[0][0]
return [[batch_size, None, self.num_classes], [batch_size, None], [batch_size, ], [batch_size, None, 4],
[batch_size, None]]
def get_config(self):
"""
Gets the configuration of this layer.
Returns
Dictionary containing the parameters of this layer.
"""
config = super(FSAFTarget, self).get_config()
config.update({'num_classes': self.num_classes})
return config
class Locations(Layer):
"""
Keras layer for generating anchors for a given shape.
"""
def __init__(self, strides, *args, **kwargs):
"""
Initializer for an Anchors layer.
Args
strides: The strides mapping to the feature maps.
"""
self.strides = strides
super(Locations, self).__init__(**kwargs)
def call(self, inputs, **kwargs):
features = inputs
feature_shapes = [tf.shape(feature)[1:3] for feature in features]
locations_per_feature = []
strides_per_feature = []
for feature_shape, stride in zip(feature_shapes, self.strides):
fh = feature_shape[0]
fw = feature_shape[1]
shifts_x = K.arange(0, fw * stride, step=stride, dtype=tf.float32)
shifts_y = K.arange(0, fh * stride, step=stride, dtype=tf.float32)
shift_x, shift_y = tf.meshgrid(shifts_x, shifts_y)
shift_x = K.reshape(shift_x, (-1,))
shift_y = K.reshape(shift_y, (-1,))
locations = K.stack((shift_x, shift_y), axis=1) + stride // 2
locations_per_feature.append(locations)
strides = tf.ones((fh, fw)) * stride
strides = tf.reshape(strides, (-1,))
strides_per_feature.append(strides)
locations = K.concatenate(locations_per_feature, axis=0)
locations = tf.tile(tf.expand_dims(locations, axis=0), (tf.shape(inputs[0])[0], 1, 1))
strides = tf.concat(strides_per_feature, axis=0)
strides = tf.tile(tf.expand_dims(strides, axis=0), (tf.shape(inputs[0])[0], 1))
return [locations, strides]
def compute_output_shape(self, input_shapes):
feature_shapes = [feature_shape[1:3] for feature_shape in input_shapes]
total = 1
for feature_shape in feature_shapes:
if None not in feature_shape:
total = total * feature_shape[0] * feature_shape[1]
else:
return [[input_shapes[0][0], None, 2], [input_shapes[0][0], None]]
return [[input_shapes[0][0], total, 2], [input_shapes[0][0], total]]
def get_config(self):
base_config = super(Locations, self).get_config()
base_config.update({'strides': self.strides})
return base_config
class RegressBoxes(Layer):
"""
Keras layer for applying regression values to boxes.
"""
def __init__(self, *args, **kwargs):
"""
Initializer for the RegressBoxes layer.
"""
super(RegressBoxes, self).__init__(**kwargs)
def call(self, inputs, **kwargs):
locations, strides, regression = inputs
x1 = locations[:, :, 0] - regression[:, :, 0] * 4.0
y1 = locations[:, :, 1] - regression[:, :, 1] * 4.0
x2 = locations[:, :, 0] + regression[:, :, 2] * 4.0
y2 = locations[:, :, 1] + regression[:, :, 3] * 4.0
bboxes = K.stack([x1, y1, x2, y2], axis=-1)
return bboxes
def compute_output_shape(self, input_shape):
return input_shape[2]
def get_config(self):
base_config = super(RegressBoxes, self).get_config()
return base_config