def input_fn():
with tf.name_scope('post_forward'):
out_shape = [FLAGS.train_image_size] * 2
anchor_creator = anchor_manipulator.AnchorCreator(
out_shape,
layers_shapes=[(38, 38), (19, 19), (10, 10), (5, 5), (3, 3),
(1, 1)],
anchor_scales=[(0.1, ), (0.2, ), (0.375, ), (0.55, ),
(0.725, ), (0.9, )],
extra_anchor_scales=[(0.1414, ), (0.2739, ), (0.4541, ),
(0.6315, ), (0.8078, ), (0.9836, )],
anchor_ratios=[(1., 2., .5), (1., 2., 3., .5, 0.3333),
(1., 2., 3., .5, 0.3333),
(1., 2., 3., .5, 0.3333), (1., 2., .5),
(1., 2., .5)],
layer_steps=[8, 16, 32, 64, 100, 300])
all_anchors, all_num_anchors_depth, all_num_anchors_spatial = anchor_creator.get_all_anchors(
)
num_anchors_per_layer = []
for ind in range(len(all_anchors)):
num_anchors_per_layer.append(all_num_anchors_depth[ind] *
all_num_anchors_spatial[ind])
anchor_encoder_decoder = anchor_manipulator.AnchorEncoder(
allowed_borders=[1.0] * 6,
positive_threshold=FLAGS.match_threshold,
ignore_threshold=FLAGS.neg_threshold,
prior_scaling=[0.1, 0.1, 0.2, 0.2])
# global global_anchor_info
# global_anchor_info = {'decode_fn': lambda pred : anchor_encoder_decoder.decode_all_anchors(pred, num_anchors_per_layer),
# 'num_anchors_per_layer': num_anchors_per_layer,
# 'all_num_anchors_depth': all_num_anchors_depth,
# 'encode_fn': lambda glabels_, gbboxes_: anchor_encoder_decoder.encode_all_anchors(glabels_, gbboxes_, all_anchors, all_num_anchors_depth, all_num_anchors_spatial)}
image_preprocessing_fn = lambda image_, labels_, bboxes_: ssd_preprocessing.preprocess_image(
image_,
labels_,
bboxes_,
out_shape,
is_training=is_training,
data_format=FLAGS.data_format,
output_rgb=False)
# anchor_encoder_fn = lambda glabels_, gbboxes_: anchor_encoder_decoder.encode_all_anchors(glabels_, gbboxes_, all_anchors, all_num_anchors_depth, all_num_anchors_spatial)
filenames = tf.placeholder(tf.string, shape=[None])
arga = tf.constant(True)
dataset = tf.data.TFRecordDataset(training_files)
dataset = dataset.map(lambda x: data_mapping_fn(
x, is_training, image_preprocessing_fn))
dataset = dataset.repeat() # repeat the input infinitely
dataset = dataset.batch(batch_size) # set the batch size
iterator = dataset.make_initializable_iterator()
# return image, {'shape': shape, 'loc_targets': loc_targets, 'cls_targets': cls_targets, 'match_scores': match_scores}
return dataset
def image_preprocessing_fn(image_, labels_, bboxes_):
return ssd_preprocessing.preprocess_image(image_,
labels_,
bboxes_,
out_shape,
is_training=False,
data_format=DATA_FORMAT,
output_rgb=False)
def image_preprocessing_fn(image_, labels_, bboxes_):
return ssd_preprocessing.preprocess_image(
image_,
labels_,
bboxes_,
out_shape,
is_training=is_training,
data_format=FLAGS.data_format,
output_rgb=False)
def input_fn(dataset_pattern='val-*', batch_size=1, data_location=None):
out_shape = [SSD_VGG16_IMAGE_SIZE] * 2
anchor_creator = anchor_manipulator.AnchorCreator(
out_shape,
layers_shapes=[(38, 38), (19, 19), (10, 10), (5, 5), (3, 3), (1, 1)],
anchor_scales=[(0.1, ), (0.2, ), (0.375, ), (0.55, ), (0.725, ),
(0.9, )],
extra_anchor_scales=[(0.1414, ), (0.2739, ), (0.4541, ), (0.6315, ),
(0.8078, ), (0.9836, )],
anchor_ratios=[(1., 2., .5), (1., 2., 3., .5, 0.3333),
(1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333),
(1., 2., .5), (1., 2., .5)],
layer_steps=[8, 16, 32, 64, 100, 300])
all_anchors, all_num_anchors_depth, all_num_anchors_spatial = anchor_creator.get_all_anchors(
)
num_anchors_per_layer = []
for ind in range(len(all_anchors)):
num_anchors_per_layer.append(all_num_anchors_depth[ind] *
all_num_anchors_spatial[ind])
anchor_encoder_decoder = anchor_manipulator.AnchorEncoder(
allowed_borders=[1.0] * 6,
positive_threshold=MATCH_THRESHOLD,
ignore_threshold=NEG_THRESHOLD,
prior_scaling=[0.1, 0.1, 0.2, 0.2])
image_preprocessing_fn = lambda image_, labels_, bboxes_: ssd_preprocessing.preprocess_image(
image_,
labels_,
bboxes_,
out_shape,
is_training=False,
data_format=DATA_FORMAT,
output_rgb=False)
anchor_encoder_fn = lambda glabels_, gbboxes_: anchor_encoder_decoder.encode_all_anchors(
glabels_, gbboxes_, all_anchors, all_num_anchors_depth,
all_num_anchors_spatial)
image, filename, shape, loc_targets, cls_targets, match_scores = \
dataset_common.slim_get_batch(NUM_CLASSES,
batch_size,
'val',
os.path.join(
data_location,
dataset_pattern),
NUM_READERS,
NUM_PREPROCESSING_THREADS,
image_preprocessing_fn,
anchor_encoder_fn,
num_epochs=1,
is_training=False)
return image, filename, shape
def input_fn():
out_shape = [FLAGS.train_image_size] * 2
ssd300_anchor_params = {'layers_shapes': [(38, 38), (19, 19), (10, 10), (5, 5), (3, 3), (1, 1)], 'anchor_scales': [(0.1,), (0.2,), (0.375,), (0.55,), (0.725,), (0.9,)],
'extra_anchor_scales': [(0.1414,), (0.2739,), (0.4541,), (0.6315,), (0.8078,), (0.9836,)],
'anchor_ratios': [(1., 2., .5), (1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333), (1., 2., .5), (1., 2., .5)],
'layer_steps': [8, 16, 32, 64, 100, 300]}
ssd512_anchor_params = {'layers_shapes': [(64, 64), (32, 32), (16, 16), (8, 8), (4, 4), (2, 2), (1, 1)],
'anchor_scales': [(0.07,), (0.15,), (0.3,), (0.45,), (0.6,), (0.75,), (0.9,)],
'extra_anchor_scales': [(0.1025,), (0.2121,), (0.3674,), (0.5196,), (0.6708,), (0.8216,), (0.9721,)],
'anchor_ratios': [(1., 2., .5), (1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333), (1., 2., .5), (1., 2., .5)],
'layer_steps': [8, 16, 32, 64, 128, 256, 512]}
if FLAGS.train_image_size == 512:
net_params = ssd512_anchor_params
print('using ssd512 model')
else:
net_params = ssd300_anchor_params
print('using ssd300 model')
anchor_creator = anchor_manipulator.AnchorCreator(out_shape, **net_params)
all_anchors, all_num_anchors_depth, all_num_anchors_spatial = anchor_creator.get_all_anchors()
num_anchors_per_layer = []
for ind in range(len(all_anchors)):
num_anchors_per_layer.append(all_num_anchors_depth[ind] * all_num_anchors_spatial[ind])
anchor_encoder_decoder = anchor_manipulator.AnchorEncoder(allowed_borders = [1.0] * len(net_params['layer_steps']),
positive_threshold = FLAGS.match_threshold,
ignore_threshold = FLAGS.neg_threshold,
prior_scaling=[0.1, 0.1, 0.2, 0.2])
image_preprocessing_fn = lambda image_, labels_, bboxes_ : ssd_preprocessing.preprocess_image(image_, labels_, bboxes_, out_shape, is_training=is_training, data_format=FLAGS.data_format, output_rgb=False)
anchor_encoder_fn = lambda glabels_, gbboxes_: anchor_encoder_decoder.encode_all_anchors(glabels_, gbboxes_, all_anchors, all_num_anchors_depth, all_num_anchors_spatial)
image, filename, shape, loc_targets, cls_targets, match_scores = dataset_common.slim_get_batch(FLAGS.num_classes,
batch_size,
('train' if is_training else 'val'),
os.path.join(FLAGS.data_dir, dataset_pattern),
FLAGS.num_readers,
FLAGS.num_preprocessing_threads,
image_preprocessing_fn,
anchor_encoder_fn,
num_epochs=FLAGS.train_epochs,
is_training=is_training)
global global_anchor_info
global_anchor_info = {'decode_fn': lambda pred : anchor_encoder_decoder.decode_all_anchors(pred, num_anchors_per_layer),
'num_anchors_per_layer': num_anchors_per_layer,
'all_num_anchors_depth': all_num_anchors_depth }
return {'image': image, 'filename': filename, 'shape': shape, 'loc_targets': loc_targets, 'cls_targets': cls_targets, 'match_scores': match_scores}, None
def input_fn():
assert batch_size==1, 'We only support single batch when evaluation.'
target_shape = [FLAGS.train_image_size] * 2
image_preprocessing_fn = lambda image_, labels_, bboxes_ : ssd_preprocessing.preprocess_image(image_, labels_, bboxes_, target_shape, is_training=is_training, data_format=FLAGS.data_format, output_rgb=False)
image, filename, shape, output_shape = dataset_common.slim_get_batch(FLAGS.num_classes,
batch_size,
('train' if is_training else 'val'),
os.path.join(FLAGS.data_dir, dataset_pattern),
FLAGS.num_readers,
FLAGS.num_preprocessing_threads,
image_preprocessing_fn,
None,
num_epochs=1,
is_training=is_training)
return {'image': image, 'filename': filename, 'shape': shape, 'output_shape': output_shape}, None
def input_fn():
out_shape = [FLAGS.train_image_size] * 2
anchor_creator = anchor_manipulator.AnchorCreator(out_shape,
layers_shapes = [(38, 38), (19, 19), (10, 10), (5, 5), (3, 3), (1, 1)],
anchor_scales = [(0.1,), (0.2,), (0.375,), (0.55,), (0.725,), (0.9,)],
extra_anchor_scales = [(0.1414,), (0.2739,), (0.4541,), (0.6315,), (0.8078,), (0.9836,)],
anchor_ratios = [(1., 2., .5), (1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333), (1., 2., 3., .5, 0.3333), (1., 2., .5), (1., 2., .5)],
layer_steps = [8, 16, 32, 64, 100, 300])
all_anchors, all_num_anchors_depth, all_num_anchors_spatial = anchor_creator.get_all_anchors()
# all_anchors: [[(38x38x1),(38x38x1),(4x1),(4x1)],[(19x19x1),(19x19x1),(4x1),(4x1)]... ] -> recording all the anchors information
num_anchors_per_layer = []
for ind in range(len(all_anchors)):
num_anchors_per_layer.append(all_num_anchors_depth[ind] * all_num_anchors_spatial[ind])
anchor_encoder_decoder = anchor_manipulator.AnchorEncoder(allowed_borders = [1.0] * 6,
positive_threshold = FLAGS.match_threshold,
ignore_threshold = FLAGS.neg_threshold,
prior_scaling=[0.1, 0.1, 0.2, 0.2])
image_preprocessing_fn = lambda image_, labels_, bboxes_ : ssd_preprocessing.preprocess_image(image_, labels_, bboxes_, out_shape, is_training=is_training, data_format=FLAGS.data_format, output_rgb=False)
anchor_encoder_fn = lambda glabels_, gbboxes_: anchor_encoder_decoder.encode_all_anchors(glabels_, gbboxes_, all_anchors, all_num_anchors_depth, all_num_anchors_spatial)
anchor_decoder_fn = lambda pred : anchor_encoder_decoder.decode_all_anchors(pred, num_anchors_per_layer)
image, _, shape, loc_targets, cls_targets, match_scores = dataset_common.slim_get_batch(FLAGS.num_classes,
batch_size,
('train' if is_training else 'val'),
os.path.join(FLAGS.data_dir, dataset_pattern),
FLAGS.num_readers,
FLAGS.num_preprocessing_threads,
image_preprocessing_fn,
anchor_encoder_fn,
num_epochs=FLAGS.train_epochs,
is_training=is_training)
global global_anchor_info
global_anchor_info = {'decode_fn': anchor_decoder_fn,
'num_anchors_per_layer': num_anchors_per_layer,
'all_num_anchors_depth': all_num_anchors_depth }
return image, {'shape': shape, 'loc_targets': loc_targets, 'cls_targets': cls_targets, 'match_scores': match_scores}
def slim_get_split(file_pattern='{}_????'):
# Features in Pascal VOC TFRecords.
keys_to_features = {
'image/encoded': tf.FixedLenFeature((), tf.string, default_value=''),
'image/format': tf.FixedLenFeature((), tf.string, default_value='jpeg'),
'image/height': tf.FixedLenFeature([1], tf.int64),
'image/width': tf.FixedLenFeature([1], tf.int64),
'image/channels': tf.FixedLenFeature([1], tf.int64),
'image/shape': tf.FixedLenFeature([3], tf.int64),
'image/object/bbox/xmin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/xmax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/label': tf.VarLenFeature(dtype=tf.int64),
'image/object/bbox/difficult': tf.VarLenFeature(dtype=tf.int64),
'image/object/bbox/truncated': tf.VarLenFeature(dtype=tf.int64),
}
items_to_handlers = {
'image': slim.tfexample_decoder.Image('image/encoded', 'image/format'),
'shape': slim.tfexample_decoder.Tensor('image/shape'),
'object/bbox': slim.tfexample_decoder.BoundingBox(
['ymin', 'xmin', 'ymax', 'xmax'], 'image/object/bbox/'),
'object/label': slim.tfexample_decoder.Tensor('image/object/bbox/label'),
'object/difficult': slim.tfexample_decoder.Tensor('image/object/bbox/difficult'),
'object/truncated': slim.tfexample_decoder.Tensor('image/object/bbox/truncated'),
}
decoder = slim.tfexample_decoder.TFExampleDecoder(keys_to_features, items_to_handlers)
dataset = slim.dataset.Dataset(
data_sources=file_pattern,
reader=tf.TFRecordReader,
decoder=decoder,
num_samples=100,
items_to_descriptions=None,
num_classes=21,
labels_to_names=None)
with tf.name_scope('dataset_data_provider'):
provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=2,
common_queue_capacity=32,
common_queue_min=8,
shuffle=True,
num_epochs=1)
[org_image, shape, glabels_raw, gbboxes_raw, isdifficult] = provider.get(['image', 'shape',
'object/label',
'object/bbox',
'object/difficult'])
image, glabels, gbboxes = ssd_preprocessing.preprocess_image(org_image, glabels_raw, gbboxes_raw, [300, 300],
is_training=True, data_format='channels_last',
output_rgb=True)
anchor_creator = anchor_manipulator.AnchorCreator([300] * 2,
layers_shapes=[(38, 38), (19, 19), (10, 10), (5, 5), (3, 3),
(1, 1)],
anchor_scales=[(0.1,), (0.2,), (0.375,), (0.55,), (0.725,),
(0.9,)],
extra_anchor_scales=[(0.1414,), (0.2739,), (0.4541,), (0.6315,),
(0.8078,), (0.9836,)],
anchor_ratios=[(2., .5), (2., 3., .5, 0.3333),
(2., 3., .5, 0.3333), (2., 3., .5, 0.3333),
(2., .5), (2., .5)],
layer_steps=[8, 16, 32, 64, 100, 300])
all_anchors, all_num_anchors_depth, all_num_anchors_spatial = anchor_creator.get_all_anchors()
num_anchors_per_layer = []
for ind in range(len(all_anchors)):
num_anchors_per_layer.append(all_num_anchors_depth[ind] * all_num_anchors_spatial[ind])
anchor_encoder_decoder = anchor_manipulator.AnchorEncoder(allowed_borders=[1.0] * 6,
positive_threshold=0.5,
ignore_threshold=0.5,
prior_scaling=[0.1, 0.1, 0.2, 0.2])
gt_targets, gt_labels, gt_scores = anchor_encoder_decoder.encode_all_anchors(glabels, gbboxes, all_anchors,
all_num_anchors_depth,
all_num_anchors_spatial, True)
anchors = anchor_encoder_decoder._all_anchors
# split by layers
gt_targets, gt_labels, gt_scores, anchors = tf.split(gt_targets, num_anchors_per_layer, axis=0), \
tf.split(gt_labels, num_anchors_per_layer, axis=0), \
tf.split(gt_scores, num_anchors_per_layer, axis=0), \
[tf.split(anchor, num_anchors_per_layer, axis=0) for anchor in anchors]
save_image_op = tf.py_func(save_image_with_bbox,
[ssd_preprocessing.unwhiten_image(image),
tf.clip_by_value(tf.concat(gt_labels, axis=0), 0, tf.int64.max),
tf.concat(gt_scores, axis=0),
tf.concat(gt_targets, axis=0)],
tf.int64, stateful=True)
return save_image_op
def slim_get_split(file_pattern='{}_????'):
# Features in Pascal VOC TFRecords.
keys_to_features = {
'image/encoded': tf.FixedLenFeature((), tf.string, default_value=''),
'image/format': tf.FixedLenFeature((), tf.string,
default_value='jpeg'),
'image/height': tf.FixedLenFeature([1], tf.int64),
'image/width': tf.FixedLenFeature([1], tf.int64),
'image/channels': tf.FixedLenFeature([1], tf.int64),
'image/shape': tf.FixedLenFeature([3], tf.int64),
'image/object/bbox/xmin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymin': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/xmax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/ymax': tf.VarLenFeature(dtype=tf.float32),
'image/object/bbox/label': tf.VarLenFeature(dtype=tf.int64),
'image/object/bbox/difficult': tf.VarLenFeature(dtype=tf.int64),
'image/object/bbox/truncated': tf.VarLenFeature(dtype=tf.int64),
}
items_to_handlers = {
'image':
slim.tfexample_decoder.Image('image/encoded', 'image/format'),
'shape':
slim.tfexample_decoder.Tensor('image/shape'),
'object/bbox':
slim.tfexample_decoder.BoundingBox(['ymin', 'xmin', 'ymax', 'xmax'],
'image/object/bbox/'),
'object/label':
slim.tfexample_decoder.Tensor('image/object/bbox/label'),
'object/difficult':
slim.tfexample_decoder.Tensor('image/object/bbox/difficult'),
'object/truncated':
slim.tfexample_decoder.Tensor('image/object/bbox/truncated'),
}
decoder = slim.tfexample_decoder.TFExampleDecoder(keys_to_features,
items_to_handlers)
dataset = slim.dataset.Dataset(data_sources=file_pattern,
reader=tf.TFRecordReader,
decoder=decoder,
num_samples=100,
items_to_descriptions=None,
num_classes=21,
labels_to_names=None)
with tf.name_scope('dataset_data_provider'):
provider = slim.dataset_data_provider.DatasetDataProvider(
dataset,
num_readers=2,
common_queue_capacity=32,
common_queue_min=8,
shuffle=True,
num_epochs=1)
[org_image, shape, glabels_raw, gbboxes_raw, isdifficult] = provider.get(
['image', 'shape', 'object/label', 'object/bbox', 'object/difficult'])
image, glabels, gbboxes = ssd_preprocessing.preprocess_image(
org_image,
glabels_raw,
gbboxes_raw, [300, 300],
is_training=True,
data_format='channels_last',
output_rgb=True)
anchor_encoder_decoder = anchor_manipulator.AnchorEncoder(
positive_threshold=0.5,
ignore_threshold=0.5,
prior_scaling=[0.1, 0.1, 0.2, 0.2])
all_anchor_scales = [(30., ), (60., ), (112.5, ), (165., ), (217.5, ),
(270., )]
all_extra_scales = [(42.43, ), (82.17, ), (136.23, ), (189.45, ),
(242.34, ), (295.08, )]
all_anchor_ratios = [(2., .5), (2., 3., .5, 0.3333), (2., 3., .5, 0.3333),
(2., 3., .5, 0.3333), (2., .5), (2., .5)]
all_layer_shapes = [(38, 38), (19, 19), (10, 10), (5, 5), (3, 3), (1, 1)]
all_layer_strides = [8, 16, 32, 64, 100, 300]
total_layers = len(all_layer_shapes)
anchors_height = list()
anchors_width = list()
anchors_depth = list()
for ind in range(total_layers):
_anchors_height, _anchors_width, _anchor_depth = anchor_encoder_decoder.get_anchors_width_height(
all_anchor_scales[ind], all_extra_scales[ind],
all_anchor_ratios[ind])
anchors_height.append(_anchors_height)
anchors_width.append(_anchors_width)
anchors_depth.append(_anchor_depth)
anchors_ymin, anchors_xmin, anchors_ymax, anchors_xmax, inside_mask = anchor_encoder_decoder.get_all_anchors(
[300] * 2, anchors_height, anchors_width, anchors_depth,
[0.5] * total_layers, all_layer_shapes, all_layer_strides,
[300.] * total_layers, [False] * total_layers)
gt_targets, gt_labels, gt_scores = anchor_encoder_decoder.encode_anchors(
glabels, gbboxes, anchors_ymin, anchors_xmin, anchors_ymax,
anchors_xmax, inside_mask, True)
num_anchors_per_layer = list()
for ind, layer_shape in enumerate(all_layer_shapes):
_, _num_anchors_per_layer = anchor_encoder_decoder.get_anchors_count(
anchors_depth[ind], layer_shape)
num_anchors_per_layer.append(_num_anchors_per_layer)
# split by layers
all_anchors = tf.stack(
[anchors_ymin, anchors_xmin, anchors_ymax, anchors_xmax], axis=-1)
gt_targets, gt_labels, gt_scores, anchors = tf.split(gt_targets, num_anchors_per_layer, axis=0),\
tf.split(gt_labels, num_anchors_per_layer, axis=0),\
tf.split(gt_scores, num_anchors_per_layer, axis=0),\
tf.split(all_anchors, num_anchors_per_layer, axis=0)
save_image_op = tf.py_func(save_image_with_bbox, [
ssd_preprocessing.unwhiten_image(image),
tf.clip_by_value(tf.concat(gt_labels, axis=0), 0, tf.int64.max),
tf.concat(gt_scores, axis=0),
tf.concat(gt_targets, axis=0)
],
tf.int64,
stateful=True)
return save_image_op
def input_fn():
#train_imgage_size = 300 [300, 300]
target_shape = [FLAGS.train_image_size] * 2
#match_threshold:0.5
#neg_threshold:0.5
anchor_encoder_decoder = anchor_manipulator.AnchorEncoder(
positive_threshold=FLAGS.match_threshold,
ignore_threshold=FLAGS.neg_threshold,
prior_scaling=[0.1, 0.1, 0.2, 0.2])
all_anchor_scales = [(30., ), (60., ), (112.5, ), (165., ), (217.5, ),
(270., )]
all_extra_scales = [(42.43, ), (82.17, ), (136.23, ), (189.45, ),
(242.34, ), (295.08, )]
all_anchor_ratios = [(1., 2., .5), (1., 2., 3., .5, 0.3333),
(1., 2., 3., .5, 0.3333),
(1., 2., 3., .5, 0.3333), (1., 2., .5),
(1., 2., .5)]
all_layer_shapes = [(38, 38), (19, 19), (10, 10), (5, 5), (3, 3),
(1, 1)]
all_layer_strides = [8, 16, 32, 64, 100, 300]
total_layers = len(all_layer_shapes)
anchors_height = list()
anchors_width = list()
anchors_depth = list()
for ind in range(total_layers):
#若该层有n个default_prior_box则anchors_height是这些box的h，_anchor_depth是n
_anchors_height, _anchors_width, _anchor_depth = anchor_encoder_decoder.get_anchors_width_height(
all_anchor_scales[ind],
all_extra_scales[ind],
all_anchor_ratios[ind],
name='get_anchors_width_height{}'.format(ind))
anchors_height.append(_anchors_height)
anchors_width.append(_anchors_width)
anchors_depth.append(_anchor_depth)
#anchors_ymin: [38*38*4 + 19*19*6 + 10*10*6 + 5*5*6 + 3*3*4 + 1*!*4]
anchors_ymin, anchors_xmin, anchors_ymax, anchors_xmax, inside_mask = anchor_encoder_decoder.get_all_anchors(
target_shape, anchors_height, anchors_width, anchors_depth,
[0.5] * total_layers, all_layer_shapes, all_layer_strides,
[FLAGS.train_image_size * 1.] * total_layers,
[False] * total_layers)
num_anchors_per_layer = list()
#all_layer_shapes [(38, 38), (19, 19), (10, 10), (5, 5), (3, 3), (1, 1)]
for ind, layer_shape in enumerate(all_layer_shapes):
#num_anchors_per_layer:layer_shape[0]*layer_layer[1]*anchors_depth
_, _num_anchors_per_layer = anchor_encoder_decoder.get_anchors_count(
anchors_depth[ind],
layer_shape,
name='get_anchor_count{}'.format(ind))
num_anchors_per_layer.append(_num_anchors_per_layer)
#num_anchors_per_layer:[38*38*4, 19*19*6, 10*10*6, 5*5*6, 3*3*4, 1*!*4]
image_preprocessing_fn = lambda image_, labels_, bboxes_: ssd_preprocessing.preprocess_image(
image_,
labels_,
bboxes_,
target_shape,
is_training=is_training,
data_format=FLAGS.data_format,
output_rgb=False)
anchor_encoder_fn = lambda glabels_, gbboxes_: anchor_encoder_decoder.encode_anchors(
glabels_, gbboxes_, anchors_ymin, anchors_xmin, anchors_ymax,
anchors_xmax, inside_mask)
image, _, shape, loc_targets, cls_targets, match_scores = dataset_common.slim_get_batch(
FLAGS.num_classes,
batch_size, ('train' if is_training else 'val'),
os.path.join(FLAGS.data_dir, dataset_pattern),
FLAGS.num_readers,
FLAGS.num_preprocessing_threads,
image_preprocessing_fn,
anchor_encoder_fn,
num_epochs=FLAGS.train_epochs,
is_training=is_training)
global global_anchor_info
global_anchor_info = {
'decode_fn':
lambda pred: anchor_encoder_decoder.batch_decode_anchors(
pred, anchors_ymin, anchors_xmin, anchors_ymax, anchors_xmax),
'num_anchors_per_layer':
num_anchors_per_layer,
'all_num_anchors_depth':
anchors_depth
}
return image, {
'shape': shape,
'loc_targets': loc_targets,
'cls_targets': cls_targets,
'match_scores': match_scores
}
