def load_darknet_model(filename,
model_type,
number_of_foreground_classes,
retry_with_update_if_possible=False,
darknet_library='libdarknet.so',
darknet_cfg_filename=''):
# Extract information of anchors.
anchor_info = None
with open(darknet_cfg_filename, 'r') as fp:
line = iter(fp.readline, '')
while True:
try:
current = line.next()
except StopIteration:
break
if current.lstrip(' ').startswith('anchors'):
# Use only the line found first.
# For a YOLOv3 model, it should be fixed in future.
try:
anchor_info = [
float(s) for s in current.split('=')[1].split(',')
]
except ValueError:
print('Failed to extract numbers from the line [%s]' %
(current, ))
raise
break
#
print('anchors = %s' % (anchor_info, ))
if model_type == 'yolo_v2':
overwritten_class_variables = None
if anchor_info != None:
assert (len(anchor_info) % 2 == 0)
anchors = []
for n in range(len(anchor_info) / 2):
width = anchor_info[2 * n]
height = anchor_info[2 * n + 1]
anchors.append((height, width))
anchors = tuple(anchors)
overwritten_class_variables = {'_anchors': anchors}
print('anchors: %s' % (anchors, ))
model = make_serializable_object(
chainercv.links.YOLOv2,
constructor_args={
'n_fg_class': number_of_foreground_classes,
},
overwritten_class_variables=overwritten_class_variables)
elif model_type == 'yolo_v3':
overwritten_class_variables = None
if anchor_info != None:
# Currently, the anchor information is generated from
# only the first line that includes 'anchors = ' in the
# cfg file.
# It may have to be fixed in future.
assert (len(anchor_info) % (2 * 3) == 0)
anchors = []
offset = 0
number_of_pairs = len(anchor_info) / 2
number_in_group = number_of_pairs / 3
for l in range(3):
anchors_in_group = []
offset = l * number_in_group * 2
for n in range(number_in_group):
width = int(anchor_info[offset + 2 * n])
height = int(anchor_info[offset + 2 * n + 1])
anchors_in_group.append((height, width))
anchors.append(tuple(anchors_in_group))
anchors = tuple(reversed(anchors))
overwritten_class_variables = {'_anchors': anchors}
print('anchors: %s' % (anchors, ))
model = make_serializable_object(
chainercv.links.YOLOv3,
constructor_args={
'n_fg_class': number_of_foreground_classes,
},
overwritten_class_variables=overwritten_class_variables)
with chainer.using_config('train', False):
model(np.empty((1, 3, model.insize, model.insize), dtype=np.float32))
model_is_old_format = False
with open(filename, mode='rb') as f:
major = np.fromfile(f, dtype=np.int32, count=1)
minor = np.fromfile(f, dtype=np.int32, count=1)
np.fromfile(f, dtype=np.int32, count=1) # revision
if major == 0 and minor <= 1:
model_is_old_format = True
print(
'The file "%s" is written in an old format, where (major, minor) == (%d, %d) '
% (filename, major, minor))
sys.stdout.flush()
model = None
else:
print(
'The file "%s" is written in a new format, where (major, minor) == (%d, %d) '
% (filename, major, minor))
assert (major * 10 + minor >= 2 and major < 1000 and minor < 1000)
model_is_old_format = False
np.fromfile(f, dtype=np.int64, count=1) # seen
if model_type == 'yolo_v2':
load_yolo_v2(f, model)
elif model_type == 'yolo_v3':
load_yolo_v3(f, model)
if model_is_old_format and retry_with_update_if_possible:
import tempfile
import os
print('Try to convert "%s" to the new format and load it.' %
(filename, ))
sys.stdout.flush()
fileno, tmp_path = tempfile.mkstemp()
try:
succeeded = update_format_of_darknet_model(tmp_path, filename,
darknet_library,
darknet_cfg_filename)
if succeeded:
print(
'Succeeded to update "%s" and save the updated version into "%s".'
% (filename, tmp_path))
sys.stdout.flush()
model = load_darknet_model(
tmp_path,
model_type,
number_of_foreground_classes,
retry_with_update_if_possible=False,
darknet_cfg_filename=darknet_cfg_filename)
else:
print('Failed to convert "%s".' % (filename, ))
sys.stdout.flush()
finally:
# remove the temporary file
os.remove(tmp_path)
return model
import rospy
parsed_args = rospy.myargv(argv=sys.argv)
except ImportError:
pass
args = parser.parse_args(parsed_args[1:])
print("Preparing the pretrained model...")
sys.stdout.flush()
if args.model == 'faster_rcnn':
model = make_serializable_object(
chainercv.links.FasterRCNNVGG16,
constructor_args={
'n_fg_class': len(voc_bbox_label_names),
'pretrained_model': args.pretrained_model,
},
template_args={
# Do not retrieve the pre-trained model again on generating a
# template object for loading weights in a file.
'n_fg_class': len(voc_bbox_label_names),
'pretrained_model': None,
})
elif args.model == 'yolo_v2':
model = make_serializable_object(
chainercv.links.YOLOv2,
constructor_args={
'n_fg_class': len(voc_bbox_label_names),
'pretrained_model': args.pretrained_model,
},
template_args={
# Do not retrieve the pre-trained model again on generating a
# template object for loading weights in a file.