def __init__(self):
super().__init__()
with self.init_scope():
# bottom up
self.resnet = ResNet50Layers('auto')
del self.resnet.fc6
# top layer (reduce channel)
self.toplayer = L.Convolution2D(
in_channels=None, out_channels=256, ksize=1, stride=1, pad=0)
# conv layer for top-down pathway
self.conv_p4 = L.Convolution2D(None, 256, ksize=3, stride=1, pad=1)
self.conv_p3 = L.Convolution2D(None, 256, ksize=3, stride=1, pad=1)
self.conv_p2 = L.Convolution2D(None, 256, ksize=3, stride=1, pad=1)
self.conv_p6 = L.Convolution2D(None, 256, ksize=1, stride=2, pad=0)
# lateral connection
self.lat_p4 = L.Convolution2D(
in_channels=None, out_channels=256, ksize=1, stride=1, pad=0)
self.lat_p3 = L.Convolution2D(
in_channels=None, out_channels=256, ksize=1, stride=1, pad=0)
self.lat_p2 = L.Convolution2D(
in_channels=None, out_channels=256, ksize=1, stride=1, pad=0)
# anchor_sizes / anchor_base anchor_base is invisible from lamba function??
self.anchor_scales = list(
map(lambda x: x/float(self.anchor_base), self.anchor_sizes))
def create_model(weight='auto', activate=F.sigmoid):
resnet = ResNet50Layers(pretrained_model=weight)
model = chainer.Sequential(lambda x: resnet(x, layers=['res5'])['res5'],
L.Linear(None, 128))
if activate:
model.append(activate)
return model
def _copy_imagenet_pretrained_resnet(self):
if self._n_layers == 50:
pretrained_model = ResNet50Layers(pretrained_model='auto')
elif self._n_layers == 101:
pretrained_model = ResNet101Layers(pretrained_model='auto')
else:
raise ValueError
self.extractor.conv1.copyparams(pretrained_model.conv1)
# The pretrained weights are trained to accept BGR images.
# Convert weights so that they accept RGB images.
self.extractor.conv1.W.data[:] = \
self.extractor.conv1.W.data[:, ::-1]
self.extractor.bn1.copyparams(pretrained_model.bn1)
_copy_persistent_chain(self.extractor.bn1, pretrained_model.bn1)
self.extractor.res2.copyparams(pretrained_model.res2)
_copy_persistent_chain(self.extractor.res2, pretrained_model.res2)
self.extractor.res3.copyparams(pretrained_model.res3)
_copy_persistent_chain(self.extractor.res3, pretrained_model.res3)
self.extractor.res4.copyparams(pretrained_model.res4)
_copy_persistent_chain(self.extractor.res4, pretrained_model.res4)
self.head.res5.copyparams(pretrained_model.res5)
_copy_persistent_chain(self.head.res5, pretrained_model.res5)
def __init__(self, initialW=None):
super().__init__()
with self.init_scope():
# bottom up
self.resnet = ResNet50Layers('auto')
del self.resnet.fc6
# top layer (reduce channel)
self.toplayer = L.Convolution2D(in_channels=None,
out_channels=256,
ksize=1,
stride=1,
pad=0,
initialW=initialW)
# conv layer for top-down pathway
self.conv_p4 = L.Convolution2D(None,
256,
ksize=3,
stride=1,
pad=1,
initialW=initialW)
self.conv_p3 = L.Convolution2D(None,
256,
ksize=3,
stride=1,
pad=1,
initialW=initialW)
self.conv_p2 = L.Convolution2D(None,
256,
ksize=3,
stride=1,
pad=1,
initialW=initialW)
self.conv_p7 = L.Convolution2D(None,
256,
ksize=3,
stride=1,
pad=1,
initialW=initialW)
# lateral connection
self.lat_p4 = L.Convolution2D(in_channels=None,
out_channels=256,
ksize=1,
stride=1,
pad=0,
initialW=initialW)
self.lat_p3 = L.Convolution2D(in_channels=None,
out_channels=256,
ksize=1,
stride=1,
pad=0,
initialW=initialW)
self.lat_p2 = L.Convolution2D(in_channels=None,
out_channels=256,
ksize=1,
stride=1,
pad=0,
initialW=initialW)
def __init__(self, activate, init_scale, dropout_ratio=0.5, weight='auto'):
super().__init__()
with self.init_scope():
self.resnet = ResNet50Layers(weight)
self.fc = L.Linear(
None,
128,
initialW=chainer.initializers.LeCunNormal(scale=init_scale))
self.activate = activate
self.dropout_ratio = dropout_ratio
def main():
parser = argparse.ArgumentParser(description='evaluate imagenet')
parser.add_argument('--gpu', type=int, default=-1)
parser.add_argument('--count-by', choices=(None, 'layers', 'functions'),
default=None)
parser.add_argument('model',
choices=('vgg16', 'googlenet',
'resnet50', 'resnet101', 'resnet152'))
args = parser.parse_args()
if args.model == 'vgg16':
model = VGG16Layers(pretrained_model=None)
elif args.model == 'googlenet':
model = GoogLeNet(pretrained_model=None)
elif args.model == 'resnet50':
model = ResNet50Layers(pretrained_model=None)
elif args.model == 'resnet101':
model = ResNet101Layers(pretrained_model=None)
elif args.model == 'resnet152':
model = ResNet152Layers(pretrained_model=None)
# override batch_normalization, don't override in prodution
if 'resnet' in args.model:
monkey.override_bn()
if args.gpu >= 0:
chainer.cuda.get_device(args.gpu).use()
model.to_gpu(args.gpu)
else:
if chainer.config.use_ideep != "never":
model.to_intel64()
if args.count_by is not None and args.gpu > 0:
warnings.warn("count_by with GPU is not supported", ValueError)
if args.count_by == 'layers':
from perf_counter import Counter
monkey.decorate_link(model, Counter)
if args.count_by == 'functions':
from perf_counter import CounterHook
else:
from monkey import CounterHook
image = np.zeros((1, 3, 224, 224), dtype=np.float32) # dummy image
with chainer.using_config('train', False):
with chainer.using_config('enable_backprop', False):
with CounterHook() as counter:
model.predict(image, oversample=False)
for fn, float_ops in counter.call_history:
print('"{}","{}"'.format(fn, float_ops))
def __init__(self,
n_class,
roi_size,
spatial_scale,
loc_initialW=None,
score_initialW=None,
mask_initialW=None):
# n_class includes the background
super().__init__()
with self.init_scope():
# res5ブロックがほしいだけなのに全部読み込むのは無駄ではある
resnet50 = ResNet50Layers()
self.res5 = copy.deepcopy(resnet50.res5)
# strideは1にする
self.res5.a.conv1.stride = (1, 1)
self.res5.a.conv4.stride = (1, 1)
# 論文 図3の左から2つめ
self.conv1 = L.Convolution2D(
in_channels=None, out_channels=2048, ksize=3, stride=1, pad=1)
# マスク推定ブランチへ
self.deconv1 = L.Deconvolution2D(
in_channels=None,
out_channels=256,
ksize=2,
stride=2,
pad=0,
initialW=mask_initialW)
self.conv2 = L.Convolution2D(
in_channels=None,
out_channels=n_class - 1,
ksize=3,
stride=1,
pad=1,
initialW=mask_initialW)
self.cls_loc = L.Linear(2048, n_class * 4, initialW=loc_initialW)
self.score = L.Linear(2048, n_class, initialW=score_initialW)
self.n_class = n_class
self.roi_size = roi_size
self.spatial_scale = spatial_scale
def __init__(self, classlabels=66):
super(Encoder, self).__init__(
#model = L.ResNet50Layers(),
model=ResNet50Layers(),
fc=L.Linear(None, classlabels))