def __init__(self, num_classes=1000):
super(FCN_32s, self).__init__()
# Load the model with convolutionalized
# fully connected layers
vgg16 = vgg.vgg16(pretrained=True, fully_conv=True)
# Copy all the feature layers as is
self.features = vgg16.features
# TODO: check if Dropout works correctly for
# fully convolutional mode
# Remove the last classification 1x1 convolution
# because it comes from imagenet 1000 class classification.
# We will perform classification on different classes
fully_conv = list(vgg16.classifier.children())
fully_conv = fully_conv[:-1]
self.fully_conv = nn.Sequential(*fully_conv)
# Get a new 1x1 convolution and randomly initialize
score_32s = nn.Conv2d(4096, num_classes, 1)
self._normal_initialization(score_32s)
self.score_32s = score_32s
def __init__(self,
nclass,
backbone='vgg16',
aux=False,
pretrained_base=True,
norm_layer=nn.BatchNorm2d,
**kwargs):
super(FCN8s, self).__init__()
self.aux = aux
if backbone == 'vgg16':
self.pretrained = vgg16(pretrained=pretrained_base).features
else:
raise RuntimeError('unknown backbone: {}'.format(backbone))
self.pool3 = nn.Sequential(*self.pretrained[:17])
self.pool4 = nn.Sequential(*self.pretrained[17:24])
self.pool5 = nn.Sequential(*self.pretrained[24:])
self.head = _FCNHead(512, nclass, norm_layer)
self.score_pool3 = nn.Conv2d(256, nclass, 1)
self.score_pool4 = nn.Conv2d(512, nclass, 1)
if aux:
self.auxlayer = _FCNHead(512, nclass, norm_layer)
self.__setattr__('exclusive',
['head', 'score_pool3', 'score_pool4', 'auxlayer']
if aux else ['head', 'score_pool3', 'score_pool4'])
def __init__(self, load_weights=False):
super(CSRNet, self).__init__()
self.seen = 0
self.frontend_feat = [
64,
64,
"M",
128,
128,
"M",
256,
256,
256,
"M",
512,
512,
512,
]
self.backend_feat = [512, 512, 512, 256, 128, 64]
self.frontend = make_layers(self.frontend_feat)
self.backend = make_layers(self.backend_feat,
in_channels=512,
dilation=True)
self.output_layer = nn.Conv2d(64, 1, kernel_size=1)
if not load_weights:
mod = vgg16(pretrained=True)
pretrain_models = flow.load(
"vgg_imagenet_pretrain_model/vgg16_oneflow_model")
mod.load_state_dict(pretrain_models)
self._initialize_weights()
for i in range(len(self.frontend.state_dict().items())):
src = list(mod.state_dict().items())[i][1]
dst = list(self.frontend.state_dict().items())[i][1].copy_(src)
def __init__(self):
self.global_step = opt.load_step
self.prepare_paths()
self.data_loader = get_loader(self.data_path, opt.b_size,
opt.scale_size, True, opt.num_workers)
self.build_model()
# self.z = Variable(torch.FloatTensor(opt.b_size, opt.h))
# self.fixed_z = Variable(torch.FloatTensor(opt.b_size, opt.h))
# self.fixed_z.data.uniform_(-1, 1)
# self.fixed_x = None
# self.criterion_perceptual = perceptual_models.PerceptualLoss(model='net-lin', net='vgg', use_gpu=True, spatial=True)
self.vgg = vgg16(pretrained=True)
self.vgg.eval()
self.criterion_l1 = nn.L1Loss()
self.criterion_l2 = nn.MSELoss()
self.adversarial_loss = torch.nn.BCELoss()
if opt.cuda:
self.set_cuda()
self.cuda = True
else:
self.cuda = False
def main():
# parse the argument
parser = argparse.ArgumentParser()
parser.add_argument(
'data_list',
help='The path of data list file, which consists of one image path per line'
)
parser.add_argument(
'model',
help='The model for image classification',
choices=[
'alexnet', 'vgg13', 'vgg16', 'vgg19', 'resnet', 'googlenet',
'inception-resnet-v2', 'inception_v4', 'xception'
])
parser.add_argument(
'params_path', help='The file which stores the parameters')
args = parser.parse_args()
# PaddlePaddle init
paddle.init(use_gpu=True, trainer_count=1)
image = paddle.layer.data(
name="image", type=paddle.data_type.dense_vector(DATA_DIM))
if args.model == 'alexnet':
out = alexnet.alexnet(image, class_dim=CLASS_DIM)
elif args.model == 'vgg13':
out = vgg.vgg13(image, class_dim=CLASS_DIM)
elif args.model == 'vgg16':
out = vgg.vgg16(image, class_dim=CLASS_DIM)
elif args.model == 'vgg19':
out = vgg.vgg19(image, class_dim=CLASS_DIM)
elif args.model == 'resnet':
out = resnet.resnet_imagenet(image, class_dim=CLASS_DIM)
elif args.model == 'googlenet':
out, _, _ = googlenet.googlenet(image, class_dim=CLASS_DIM)
elif args.model == 'inception-resnet-v2':
assert DATA_DIM == 3 * 331 * 331 or DATA_DIM == 3 * 299 * 299
out = inception_resnet_v2.inception_resnet_v2(
image, class_dim=CLASS_DIM, dropout_rate=0.5, data_dim=DATA_DIM)
elif args.model == 'inception_v4':
out = inception_v4.inception_v4(image, class_dim=CLASS_DIM)
elif args.model == 'xception':
out = xception.xception(image, class_dim=CLASS_DIM)
# load parameters
with gzip.open(args.params_path, 'r') as f:
parameters = paddle.parameters.Parameters.from_tar(f)
file_list = [line.strip() for line in open(args.data_list)]
test_data = [(paddle.image.load_and_transform(image_file, 256, 224, False)
.flatten().astype('float32'), ) for image_file in file_list]
probs = paddle.infer(
output_layer=out, parameters=parameters, input=test_data)
lab = np.argsort(-probs)
for file_name, result in zip(file_list, lab):
print "Label of %s is: %d" % (file_name, result[0])
def init_net(self):
net_args = {
"pretrained": True,
"n_input_channels": len(self.kwargs["static"]["imagery_bands"])
}
# https://pytorch.org/docs/stable/torchvision/models.html
if self.kwargs["net"] == "resnet18":
self.model = resnet.resnet18(**net_args)
elif self.kwargs["net"] == "resnet34":
self.model = resnet.resnet34(**net_args)
elif self.kwargs["net"] == "resnet50":
self.model = resnet.resnet50(**net_args)
elif self.kwargs["net"] == "resnet101":
self.model = resnet.resnet101(**net_args)
elif self.kwargs["net"] == "resnet152":
self.model = resnet.resnet152(**net_args)
elif self.kwargs["net"] == "vgg11":
self.model = vgg.vgg11(**net_args)
elif self.kwargs["net"] == "vgg11_bn":
self.model = vgg.vgg11_bn(**net_args)
elif self.kwargs["net"] == "vgg13":
self.model = vgg.vgg13(**net_args)
elif self.kwargs["net"] == "vgg13_bn":
self.model = vgg.vgg13_bn(**net_args)
elif self.kwargs["net"] == "vgg16":
self.model = vgg.vgg16(**net_args)
elif self.kwargs["net"] == "vgg16_bn":
self.model = vgg.vgg16_bn(**net_args)
elif self.kwargs["net"] == "vgg19":
self.model = vgg.vgg19(**net_args)
elif self.kwargs["net"] == "vgg19_bn":
self.model = vgg.vgg19_bn(**net_args)
else:
raise ValueError("Invalid network specified: {}".format(
self.kwargs["net"]))
# run type: 1 = fine tune, 2 = fixed feature extractor
# - replace run type option with "# of layers to fine tune"
if self.kwargs["run_type"] == 2:
layer_count = len(list(self.model.parameters()))
for layer, param in enumerate(self.model.parameters()):
if layer <= layer_count - 5:
param.requires_grad = False
# Parameters of newly constructed modules have requires_grad=True by default
# get existing number for input features
# set new number for output features to number of categories being classified
# see: https://pytorch.org/tutorials/beginner/finetuning_torchvision_models_tutorial.html
if "resnet" in self.kwargs["net"]:
num_ftrs = self.model.fc.in_features
self.model.fc = nn.Linear(num_ftrs, self.ncats)
elif "vgg" in self.kwargs["net"]:
num_ftrs = self.model.classifier[6].in_features
self.model.classifier[6] = nn.Linear(num_ftrs, self.ncats)
def __init__(self, vgg_is_fixed, retain_upto=4):
super(VGGNetFeats, self).__init__()
self.vgg_model = vgg.vgg16(pretrained=True, retain_upto=retain_upto)
self.normalizer = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
self.to_tensor = transforms.ToTensor()
self.vgg_is_fixed = vgg_is_fixed
self.retain_upto = retain_upto
print "VGGNetFeats: ", self._modules.keys()
def __init__(self):
super(EAST, self).__init__()
#param
self.TEXT_SCALE = 512
self.pi = math.pi / 1.
self.pi = mindspore.Tensor([self.pi], mindspore.float32)
#network
self.model = vgg.vgg16()
#for i = 0
self.split = P.Split(1, 2)
self.unpool0 = unpool((32, 32))
self._concat = P.Concat(axis=1)
#for i = 1
self.concat1 = P.Concat(axis=1)
self.conv1_1 = ops.conv_bn_relu(1024, 128, stride=1, kernel_size=1, padding='valid')
self.conv1_2 = ops.conv_bn_relu(128, 128, stride=1, kernel_size=3, padding='pad', padding_number=1)
self.unpool1 = unpool((64, 64))
#for i = 2
self.concat2 = P.Concat(axis=1)
self.conv2_1 = ops.conv_bn_relu(384, 64, stride=1, kernel_size=1, padding='valid')
self.conv2_2 = ops.conv_bn_relu(64, 64, stride=1, kernel_size=3, padding='pad', padding_number=1)
self.unpool2 = unpool((128, 128))
#for i = 3
self.concat3 = P.Concat(axis=1)
self.conv3_1 = ops.conv_bn_relu(192, 32, stride=1, kernel_size=1, padding='valid')
self.conv3_2 = ops.conv_bn_relu(32, 32, stride=1, kernel_size=3, padding='pad', padding_number=1)
self.conv3_3 = ops.conv_bn_relu(32, 32, stride=1, kernel_size=3, padding='pad', padding_number=1)
#output
## F_score
self.conv_for_fscore = ops._conv(32, 1, stride=1, kernel_size=1, padding='valid')
self.sigmoid_for_fscore = nn.Sigmoid()
## geo_map
self.conv_for_geo_map = ops._conv(32, 4, stride=1, kernel_size=1, padding='valid')
self.sigmoid_for_geo_map = nn.Sigmoid()
## angle_map
self.conv_for_angle_map = ops._conv(32, 1, stride=1, kernel_size=1, padding='valid')
self.sigmoid_for_angle_map = nn.Sigmoid()
## F_geometry
self.concat_for_F_geometry = P.Concat(axis=1)
## other
self.mul = P.Mul()
self.add = P.TensorAdd()
def main(argv):
art_image = cv2.imread('images/starry_night.jpg')
user_image = cv2.imread('images/trump.jpg')
image_shape = user_image.shape
image = tf.Variable(initial_value=np.random.rand(1, image_shape[0],
image_shape[1],
image_shape[2]),
dtype=tf.float32,
trainable=True,
name='output_image')
sess = tf.Session()
with tf.variable_scope('vgg'):
vgg = vgg16(image, reuse=False)
download_weights_maybe('weights/vgg16_weights.npz')
vgg.load_weights('weights/vgg16_weights.npz', sess)
style_layers = ['conv1_1', 'conv2_1', 'conv3_1', 'conv4_1', 'conv5_1']
content_layers = ['conv3_2', 'conv4_2']
feature_matrices, gram_matrices = precompute(style_layers,
content_layers,
vgg_scope='vgg',
sess=sess,
user_image=user_image,
art_image=art_image)
content_layer_ops = map(lambda layer: vgg.get_layer(layer), content_layers)
style_layer_ops = map(lambda layer: vgg.get_layer(layer), style_layers)
loss = total_loss(image, content_layer_ops, style_layer_ops,
feature_matrices, gram_matrices)
optimizer = tf.train.AdamOptimizer(
learning_rate=LEARNING_RATE).minimize(loss)
global_vars = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
sess.run(tf.variables_initializer(global_vars))
for step in xrange(550):
sys.stdout.flush()
sys.stdout.write('\r Step %i' % step)
sess.run(optimizer)
if step % 50 == 0:
print "\rLoss for step %i: %f" % (step, sess.run(loss))
cv2.imwrite('images/result.png',
sess.run(image).reshape(image_shape))
print 'Final Loss: %f' % sess.run(loss)
cv2.imwrite('images/result.png', sess.run(image).reshape(image_shape))
def __init__(self, num_classes=40, feature_path=None):
super(MyVGG, self).__init__()
net = vgg.vgg16(pretrained=True, model_path=feature_path)
net.classifier = nn.Sequential()
self.features = net
self.classifier = nn.Sequential(
nn.Linear(512 * 7 * 7, 4096),
nn.ReLU(True),
nn.Dropout(),
nn.Linear(4096, num_classes)
)
def load_model(path):
net = vgg16(pretrained=False, progress=False)
classifier = nn.Sequential(
OrderedDict([('fc1', nn.Linear(25088, 4096)), ('relu1', nn.ReLU()),
('fc2', nn.Linear(4096, 1000)), ('relu2', nn.ReLU()),
('fc3', nn.Linear(1000, 10)),
('output', nn.LogSoftmax(dim=1))]))
# 替换
net.classifier = classifier
net.to(device)
net.load_state_dict(torch.load(path))
return net
def precompute(style_layers, content_layers, vgg_scope, sess, user_image,
art_image):
image_shape = art_image.shape
x = tf.placeholder(dtype=tf.float32, shape=[1] + list(image_shape))
with tf.variable_scope(vgg_scope, reuse=True):
vgg = vgg16(x, reuse=True)
gram_matrices = []
for layer in style_layers:
layer_op = vgg.get_layer(layer)
gram_op = gram_matrix(feature_matrix(layer_op))
gram = sess.run(
gram_op, feed_dict={x: art_image.reshape([1] + list(image_shape))})
shape = list(gram.shape)
gram_matrices += [tf.constant(gram.reshape(shape[1:]))]
feature_matrices = []
if len(content_layers) > 0:
image_shape = user_image.shape
x = tf.placeholder(dtype=tf.float32, shape=[1] + list(image_shape))
with tf.variable_scope(vgg_scope, reuse=True):
vgg = vgg16(x, reuse=True)
for layer in content_layers:
layer_op = vgg.get_layer(layer)
feature_op = feature_matrix(layer_op)
feature = sess.run(
feature_op,
feed_dict={x: user_image.reshape([1] + list(image_shape))})
shape = list(feature.shape)
feature_matrices += [tf.constant(feature.reshape(shape[1:]))]
return feature_matrices, gram_matrices
def precompute_gram_matrices(image, final_endpoint='fc8'):
with tf.Session() as session:
end_points = vgg.vgg16(image, final_endpoint=final_endpoint)
# In vgg file, we have tf.variable_scope('vgg_16', [inputs], reuse=reuse) as sc:
# In vgg file, we have variable scope named 'vgg_16'
tf.train.Saver(slim.get_variable('vgg_16')).restore(
session, vgg.checkpoint_file)
# The gram_matrix function will be defined later
# The end_points is in the vgg.py file
# it contains the layers' name (like conv1) and values.
return dict([(key, gram_matrix(value).eval())
for key, value in end_points.iteritems()])
def load_model(args):
if args.model == 'vgg11':
model = vgg.vgg11().to(device)
if args.model == 'vgg13':
model = vgg.vgg13().to(device)
if args.model == 'vgg16':
model = vgg.vgg16().to(device)
elif args.model == 'vgg19':
model = vgg.vgg19().to(device)
elif args.model == 'modified_vgg11':
model = modified_vgg.vgg11().to(device)
elif args.model == 'modified_vgg13':
model = modified_vgg.vgg13().to(device)
elif args.model == 'modified_vgg16':
model = modified_vgg.vgg16().to(device)
elif args.model == 'modified_vgg19':
model = modified_vgg.vgg19().to(device)
return model
def __init__(self,
nclass,
backbone='vgg16',
aux=False,
pretrained_base=True,
norm_layer=nn.BatchNorm2d,
**kwargs):
super(FCN32s, self).__init__()
self.aux = aux
if backbone == 'vgg16':
self.pretrained = vgg16(pretrained=pretrained_base).features
else:
raise RuntimeError('unknown backbone: {}'.format(backbone))
self.head = _FCNHead(512, nclass, norm_layer)
if aux:
self.auxlayer = _FCNHead(512, nclass, norm_layer)
self.__setattr__('exclusive',
['head', 'auxlayer'] if aux else ['head'])
def main(_):
x, img = load_image(FLAGS.input)
###x.shape (1, 224, 224, 3)
###img.shape (224, 224, 3)
sess = tf.Session()
print("\nLoading Vgg")
imgs = tf.placeholder(tf.float32, [None, 224, 224, 3])
vgg = vgg16(imgs, 'vgg.npz', sess)
print("\nFeedforwarding")
prob = sess.run(vgg.probs, feed_dict={vgg.imgs: x})[0]#probs是网络中fc3输出的结果
preds = (np.argsort(prob)[::-1])[0:5]
#argsort函数返回的是数组值从小到大的索引值,[::-1]倒顺序
print('\nTop 5 classes are')
for p in preds:
print(class_names[p], prob[p])
# Target class
predicted_class = preds[0]
# Target layer for visualization
layer_name = FLAGS.layer_name
# Number of output classes of model being used
nb_classes = 3
cam3 = grad_cam(x, vgg, sess, predicted_class, layer_name, nb_classes)
np.save('cam.npy',cam3)
img = img.astype(float)
img /= img.max()
# Superimposing the visualization with the image.
#将可视化与图像叠加在一起
new_img = img+3*cam3
new_img /= new_img.max()
# Display and save
io.imshow(new_img)
plt.show()
io.imsave(FLAGS.output, new_img)
print('-'*19,'运行完毕','-'*19)
def __init__(self, num_classes): super().__init__() feats = list(vgg.vgg16(pretrained=True).features.children()) self.feats = nn.Sequential(*feats[0:10]) self.feat3 = nn.Sequential(*feats[10:17]) self.feat4 = nn.Sequential(*feats[17:24]) self.feat5 = nn.Sequential(*feats[24:31]) self.fconn = nn.Sequential( nn.Conv2d(512, 4096, 7,padding=3), nn.ReLU(inplace=True), nn.Dropout(), nn.Conv2d(4096, 4096, 1), nn.ReLU(inplace=True), nn.Dropout(), ) self.score_feat3 = nn.Conv2d(256, num_classes, 1) self.score_feat4 = nn.Conv2d(512, num_classes, 1) self.score_fconn = nn.Conv2d(4096, num_classes, 1)
def __init__(self):
# hyper-parameter
self.batch_size = 256
self.lr = 1e-2
self.num_epochs = 160
# data
self.train_set = None
self.test_set = None
self.train_loader = None
self.test_loader = None
# net
# self.net = vgg16(num_classes=10).cuda()
self.net = vgg16(num_classes=10).half().cuda()
self.init_state_dict = copy.deepcopy(self.net.state_dict())
self.criterion = nn.CrossEntropyLoss()
# show
self.show_log = False
self.show_vis = False
# visdom
if self.show_vis:
self.vis = visdom.Visdom(port=8096,
env='vgg16-train-test-diff-random-seed')
# save date
self.train_batch_loss_list = []
self.train_batch_acc_list = []
self.train_epoch_loss_list = []
self.train_epoch_acc_list = []
self.val_epoch_loss_list = []
self.val_epoch_acc_list = []
self.params_l2_norm_list = []
def main(_):
x, img = load_image(FLAGS.input)
sess = tf.Session()
print("\nLoading Vgg")
imgs = tf.placeholder(tf.float32, [None, 224, 224, 3])
vgg = vgg16(imgs, 'vgg16_weights.npz', sess)
print("\nFeedforwarding")
prob = sess.run(vgg.probs, feed_dict={vgg.imgs: x})[0]
preds = (np.argsort(prob)[::-1])[0:5]
print('\nTop 5 classes are')
for p in preds:
print(class_names[p], prob[p])
# Target class
predicted_class = preds[0]
# Target layer for visualization
layer_name = FLAGS.layer_name
# Number of output classes of model being used
nb_classes = 1000
cam3 = grad_cam(x, vgg, sess, predicted_class, layer_name, nb_classes)
img = img.astype(float)
img /= img.max()
# Superimposing the visualization with the image.
new_img = img + 3 * cam3
new_img /= new_img.max()
# Display and save
io.imshow(new_img)
plt.show()
io.imsave(FLAGS.output, new_img)
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
]))
val_sampler = torch.utils.data.distributed.DistributedSampler(
val_dataset, num_replicas=hvd.size(), rank=hvd.rank())
val_loader = torch.utils.data.DataLoader(val_dataset,
batch_size=args.val_batch_size,
sampler=val_sampler,
**kwargs)
# Set up standard ResNet-50 model.
# model = models.resnet50()
# model = resnet.resnet110()
model = vgg.vgg16()
if args.cuda:
# Move model to GPU.
model.cuda()
# Horovod: scale learning rate by the number of GPUs.
# Gradient Accumulation: scale learning rate by batches_per_allreduce
optimizer = optim.SGD(model.parameters(),
lr=(args.base_lr * args.batches_per_allreduce *
hvd.size()),
momentum=args.momentum,
weight_decay=args.wd)
# Horovod: (optional) compression algorithm.
def Deeplabv3(weights='pascal_voc',
input_tensor=None,
input_shape=(512, 512, 3),
classes=21,
alpha=1.,
def_conv=False,
trainer=False):
""" Instantiates the Deeplabv3+ architecture
Optionally loads weights pre-trained
on PASCAL VOC. This model is available for TensorFlow only,
and can only be used with inputs following the TensorFlow
data format `(width, height, channels)`.
# Arguments
weights: one of 'pascal_voc' (pre-trained on pascal voc)
or None (random initialization)
input_tensor: optional Keras tensor (i.e. output of `layers.Input()`)
to use as image input for the model.
input_shape: shape of input image. format HxWxC
PASCAL VOC model was trained on (512,512,3) images
classes: number of desired classes. If classes != 21,
last layer is initialized randomly
backbone: backbone to use. one of {'xception','mobilenetv2'}
OS: determines input_shape/feature_extractor_output ratio. One of {8,16}.
Used only for xception backbone.
alpha: controls the width of the MobileNetV2 network. This is known as the
width multiplier in the MobileNetV2 paper.
- If `alpha` < 1.0, proportionally decreases the number
of filters in each layer.
- If `alpha` > 1.0, proportionally increases the number
of filters in each layer.
- If `alpha` = 1, default number of filters from the paper
are used at each layer.
Used only for mobilenetv2 backbone
# Returns
A Keras model instance.
"""
if input_tensor is None:
img_input = Input(shape=input_shape)
else:
if not K.is_keras_tensor(input_tensor):
img_input = Input(tensor=input_tensor, shape=input_shape)
else:
img_input = input_tensor
OS = 8
first_block_filters = _make_divisible(32 * alpha, 8)
x = Conv2D(first_block_filters,
kernel_size=3,
strides=(2, 2),
padding='same',
use_bias=False,
name='Conv')(img_input)
x = BatchNormalization(epsilon=1e-3, momentum=0.999, name='Conv_BN')(x)
x = Activation(relu6, name='Conv_Relu6')(x)
####### MAIN FEATURE EXTRACTOR #######
vgg_bare = vgg16(train=trainer,
num_classes=classes,
input_shape=input_shape,
deformable=def_conv,
normalizer=False,
full_model=False,
last_pooling=True)
(_, x) = vgg_bare.core(x)
######################################
# end of feature extractor
# branching for Atrous Spatial Pyramid Pooling
# Image Feature branch
#out_shape = int(np.ceil(input_shape[0] / OS))
b4 = AveragePooling2D(pool_size=(int(np.ceil(input_shape[0] / OS)),
int(np.ceil(input_shape[1] / OS))))(x)
b4 = Conv2D(256, (1, 1),
padding='same',
use_bias=False,
name='image_pooling')(b4)
b4 = BatchNormalization(name='image_pooling_BN', epsilon=1e-5)(b4)
b4 = Activation('relu')(b4)
b4 = BilinearUpsampling((int(np.ceil(input_shape[0] / OS)),
int(np.ceil(input_shape[1] / OS))))(b4)
# simple 1x1
b0 = Conv2D(256, (1, 1), padding='same', use_bias=False, name='aspp0')(x)
b0 = BatchNormalization(name='aspp0_BN', epsilon=1e-5)(b0)
b0 = Activation('relu', name='aspp0_activation')(b0)
# there are only 2 branches in mobilenetV2. not sure why
x = Concatenate()([b4, b0])
x = Conv2D(256, (1, 1),
padding='same',
use_bias=False,
name='concat_projection')(x)
x = BatchNormalization(name='concat_projection_BN', epsilon=1e-5)(x)
x = Activation('relu')(x)
x = Dropout(0.1)(x)
x = Conv2D(classes, (1, 1), padding='same',
name='logits_semantic_custom')(x)
x = BilinearUpsampling(output_size=(input_shape[0], input_shape[1]))(x)
# Ensure that the model takes into account
# any potential predecessors of `input_tensor`.
if input_tensor is not None:
inputs = get_source_inputs(input_tensor)
else:
inputs = img_input
if (trainer):
x = Flatten(name='flat')(x)
x = Dense(classes, activation='relu', name='fc')(x)
model = Model(inputs, x, name='deeplabv3plus')
# load weights
#if weights == 'pascal_voc':
# if backbone == 'xception':
# weights_path = get_file('deeplabv3_xception_tf_dim_ordering_tf_kernels.h5',
# WEIGHTS_PATH_X,
# cache_subdir='models')
# else:
# weights_path = get_file('deeplabv3_mobilenetv2_tf_dim_ordering_tf_kernels.h5',
# WEIGHTS_PATH_MOBILE,
# cache_subdir='models')
# model.load_weights(weights_path, by_name=True)
#elif weights == 'cityscapes':
# if backbone == 'xception':
# weights_path = get_file('deeplabv3_xception_tf_dim_ordering_tf_kernels_cityscapes.h5',
# WEIGHTS_PATH_X_CS,
# cache_subdir='models')
# else:
# weights_path = get_file('deeplabv3_mobilenetv2_tf_dim_ordering_tf_kernels_cityscapes.h5',
# WEIGHTS_PATH_MOBILE_CS,
# cache_subdir='models')
# model.load_weights(weights_path, by_name=True)
return model
import argparse
import os
import numpy as np
import cv2
import vgg
import gradcam
# Receive image path from command line
parser = argparse.ArgumentParser(description='Grad-CAM demo')
parser.add_argument('img_path', metavar='image_path', type=str, help='path to the image file')
args = parser.parse_args()
# We'll use VGG-16 for visualization
network = vgg.vgg16(pretrained=True, ctx=mx.cpu())
# We'll resize images to 224x244 as part of preprocessing
image_sz = (224, 224)
def preprocess(data):
"""Preprocess the image before running it through the network"""
data = mx.image.imresize(data, image_sz[0], image_sz[1])
data = data.astype(np.float32)
data = data/255
# These mean values were obtained from
# https://mxnet.incubator.apache.org/api/python/gluon/model_zoo.html
data = mx.image.color_normalize(data,
mean=mx.nd.array([0.485, 0.456, 0.406]),
std=mx.nd.array([0.229, 0.224, 0.225]))
data = mx.nd.transpose(data, (2,0,1)) # Channel first
return data
from torch.utils.data import DataLoader
from DatasetAB import DatasetAB
from generator import Generator
from vgg import vgg16
dataset = DatasetAB('../faces94/malestaff/tony/',
'../faces94/malestaff/voudcx/')
dataloader = DataLoader(dataset)
Attacker = Generator(180, 200, 3, 64, 256)
Victim = vgg16(num_classes=2)
print(Attacker)
print('=' * 35)
print(Victim)
for data in dataloader:
img_a, img_b = data
masked_img = Attacker(img_a, img_b)
def main_worker(gpu, ngpus_per_node, args):
perfStats = {}
for batchSize in [1, 2, 4, 8, 16, 32]:
# for batchSize in []:
args.batch_size = batchSize
args.print_freq = 500
global best_acc1
args.gpu = gpu
# create model
model = models.__dict__["vgg16"]()
if args.gpu is not None:
torch.cuda.set_device(args.gpu)
model = model.cuda(args.gpu)
else:
# DataParallel will divide and allocate batch_size to all available GPUs
if args.arch.startswith('alexnet') or args.arch.startswith('vgg'):
model.features = torch.nn.DataParallel(model.features)
model.cuda()
else:
model = torch.nn.DataParallel(model).cuda()
# define loss function (criterion) and optimizer
criterion = nn.CrossEntropyLoss().cuda(args.gpu)
optimizer = torch.optim.SGD(model.parameters(), args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
cudnn.benchmark = True
# Data loading code
traindir = os.path.join(args.data, 'train')
valdir = os.path.join(args.data, 'val')
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_dataset = SyntheticDataset((3, 224, 224), 500)
if args.distributed:
train_sampler = torch.utils.data.distributed.DistributedSampler(
train_dataset)
else:
train_sampler = None
train_loader = torch.utils.data.DataLoader(
train_dataset, batch_size=args.batch_size, shuffle=(
train_sampler is None),
num_workers=args.workers, pin_memory=True, sampler=train_sampler, drop_last=True)
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
train_sampler.set_epoch(epoch)
adjust_learning_rate(optimizer, epoch, args)
perfStats[batchSize] = Perf({0: 'load', 1: 'fp', 2: 'loss', 3: 'zero', 4: 'bp', 5: 'opt', 6: 'total/bat', 7: 'totalCPU'})
# train for one epoch
train(train_loader, model, criterion, optimizer, epoch, args, perfStats[batchSize])
if len(perfStats) == 1:
perfStats[batchSize].printHeader()
perfStats[batchSize].printAll(batchSize, 1, 0)
for batchSize in [1, 2, 4, 8, 16, 32]:
perfStats = {}
for splitCount in [1, 2, 4, 8, 16, 32]:
args.batch_size = batchSize
args.print_freq = 500
global best_acc1
args.gpu = gpu
if args.gpu is not None:
print("Use GPU: {} for training".format(args.gpu))
if args.distributed:
if args.dist_url == "env://" and args.rank == -1:
args.rank = int(os.environ["RANK"])
if args.multiprocessing_distributed:
# For multiprocessing distributed training, rank needs to be the
# global rank among all the processes
args.rank = args.rank * ngpus_per_node + gpu
dist.init_process_group(backend=args.dist_backend, init_method=args.dist_url,
world_size=args.world_size, rank=args.rank)
# create model
# if args.pretrained:
# print("=> using pre-trained model '{}'".format(args.arch))
# model = models.__dict__[args.arch](pretrained=True)
# else:
# print("=> creating model '{}'".format(args.arch))
# model = models.__dict__[args.arch]()
# if args.arch == "vgg16":
# model = vgg16()
model = vgg16(splitCount)
if args.distributed:
# For multiprocessing distributed, DistributedDataParallel constructor
# should always set the single device scope, otherwise,
# DistributedDataParallel will use all available devices.
if args.gpu is not None:
torch.cuda.set_device(args.gpu)
model.cuda(args.gpu)
# When using a single GPU per process and per
# DistributedDataParallel, we need to divide the batch size
# ourselves based on the total number of GPUs we have
args.batch_size = int(args.batch_size / ngpus_per_node)
model = torch.nn.parallel.DistributedDataParallel(
model, device_ids=[args.gpu])
else:
model.cuda()
# DistributedDataParallel will divide and allocate batch_size to all
# available GPUs if device_ids are not set
model = torch.nn.parallel.DistributedDataParallel(model)
elif args.gpu is not None:
torch.cuda.set_device(args.gpu)
model = model.cuda(args.gpu)
else:
# DataParallel will divide and allocate batch_size to all available GPUs
if args.arch.startswith('alexnet') or args.arch.startswith('vgg'):
model.features = torch.nn.DataParallel(model.features)
model.cuda()
else:
model = torch.nn.DataParallel(model).cuda()
# define loss function (criterion) and optimizer
criterion = nn.CrossEntropyLoss().cuda(args.gpu)
optimizer = torch.optim.SGD(model.parameters(), args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
cudnn.benchmark = True
# Data loading code
traindir = os.path.join(args.data, 'train')
valdir = os.path.join(args.data, 'val')
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
# train_dataset = datasets.ImageFolder(
# traindir,
# transforms.Compose([
# transforms.RandomResizedCrop(224),
# transforms.RandomHorizontalFlip(),
# transforms.ToTensor(),
# normalize,
# ]))
train_dataset = SyntheticDataset((3, 224, 224), 500)
if args.distributed:
train_sampler = torch.utils.data.distributed.DistributedSampler(
train_dataset)
else:
train_sampler = None
train_loader = torch.utils.data.DataLoader(
train_dataset, batch_size=args.batch_size, shuffle=(
train_sampler is None),
num_workers=args.workers, pin_memory=True, sampler=train_sampler, drop_last=True)
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
train_sampler.set_epoch(epoch)
adjust_learning_rate(optimizer, epoch, args)
perfStats[splitCount] = Perf({0: 'load', 1: 'fp', 2: 'loss', 3: 'zero', 4: 'bp', 5: 'opt', 6: 'total/bat', 7: 'totalCPU'})
# train for one epoch
train(train_loader, model, criterion, optimizer, epoch, args, perfStats[splitCount])
if len(perfStats) == 1:
perfStats[splitCount].printHeader()
perfStats[splitCount].printAll(batchSize, splitCount, 0)
neptune.log_text('model', model_name)
neptune.log_text('date_time', date_time)
neptune.create_experiment(model_name)
NeptuneLog()
if model_name == 'vgg11':
model = vgg.vgg11(pretrained=pretrain_check)
elif model_name == 'vgg11_bn':
model = vgg.vgg11_bn(pretrained=pretrain_check)
elif model_name == 'vgg13':
model = vgg.vgg13(pretrained=pretrain_check)
elif model_name == 'vgg13_bn':
model = vgg.vgg13_bn(pretrained=pretrain_check)
elif model_name == 'vgg16':
model = vgg.vgg16(pretrained=pretrain_check)
elif model_name == 'vgg16_bn':
model = vgg.vgg16_bn(pretrained=pretrain_check)
elif model_name == 'vgg19':
model = vgg.vgg19(pretrained=pretrain_check)
elif model_name == 'vgg19_bn':
model = vgg.vgg19_bn(pretrained=pretrain_check)
model.eval()
model = torch.nn.DataParallel(model).cuda()
optimizer = optim.Adam(model.parameters(),
lr=learning_rate,
betas=(0.9, 0.999),
eps=1e-08,
weight_decay=1e-5)
scheduler = ReduceLROnPlateau(optimizer,
def main():
# parse the argument
parser = argparse.ArgumentParser()
parser.add_argument(
'model',
help='The model for image classification',
choices=['alexnet', 'vgg13', 'vgg16', 'vgg19', 'resnet', 'googlenet'])
args = parser.parse_args()
# PaddlePaddle init
paddle.init(use_gpu=True, trainer_count=7)
image = paddle.layer.data(
name="image", type=paddle.data_type.dense_vector(DATA_DIM))
lbl = paddle.layer.data(
name="label", type=paddle.data_type.integer_value(CLASS_DIM))
extra_layers = None
learning_rate = 0.01
if args.model == 'alexnet':
out = alexnet.alexnet(image, class_dim=CLASS_DIM)
elif args.model == 'vgg13':
out = vgg.vgg13(image, class_dim=CLASS_DIM)
elif args.model == 'vgg16':
out = vgg.vgg16(image, class_dim=CLASS_DIM)
elif args.model == 'vgg19':
out = vgg.vgg19(image, class_dim=CLASS_DIM)
elif args.model == 'resnet':
out = resnet.resnet_imagenet(image, class_dim=CLASS_DIM)
learning_rate = 0.1
elif args.model == 'googlenet':
out, out1, out2 = googlenet.googlenet(image, class_dim=CLASS_DIM)
loss1 = paddle.layer.cross_entropy_cost(
input=out1, label=lbl, coeff=0.3)
paddle.evaluator.classification_error(input=out1, label=lbl)
loss2 = paddle.layer.cross_entropy_cost(
input=out2, label=lbl, coeff=0.3)
paddle.evaluator.classification_error(input=out2, label=lbl)
extra_layers = [loss1, loss2]
cost = paddle.layer.classification_cost(input=out, label=lbl)
# Create parameters
parameters = paddle.parameters.create(cost)
# Create optimizer
optimizer = paddle.optimizer.Momentum(
momentum=0.9,
regularization=paddle.optimizer.L2Regularization(rate=0.0005 *
BATCH_SIZE),
learning_rate=learning_rate / BATCH_SIZE,
learning_rate_decay_a=0.1,
learning_rate_decay_b=128000 * 35,
learning_rate_schedule="discexp", )
train_reader = paddle.batch(
paddle.reader.shuffle(
flowers.train(),
# To use other data, replace the above line with:
# reader.train_reader('train.list'),
buf_size=1000),
batch_size=BATCH_SIZE)
test_reader = paddle.batch(
flowers.valid(),
# To use other data, replace the above line with:
# reader.test_reader('val.list'),
batch_size=BATCH_SIZE)
# Create trainer
trainer = paddle.trainer.SGD(
cost=cost,
parameters=parameters,
update_equation=optimizer,
extra_layers=extra_layers)
# End batch and end pass event handler
def event_handler(event):
if isinstance(event, paddle.event.EndIteration):
if event.batch_id % 1 == 0:
print "\nPass %d, Batch %d, Cost %f, %s" % (
event.pass_id, event.batch_id, event.cost, event.metrics)
if isinstance(event, paddle.event.EndPass):
with gzip.open('params_pass_%d.tar.gz' % event.pass_id, 'w') as f:
trainer.save_parameter_to_tar(f)
result = trainer.test(reader=test_reader)
print "\nTest with Pass %d, %s" % (event.pass_id, result.metrics)
trainer.train(
reader=train_reader, num_passes=200, event_handler=event_handler)
import torch.nn.functional as F
import torch.optim as optim
# from model.roi_layers import ROIPool
import torchvision.ops.roi_pool as ROIPool
# from skimage import io
import vgg, cv2, time
from mydataset import MyDataset
from image_processing import resizeThermal
# from image_processing import showTensor
rgb_mean = (0.4914, 0.4822, 0.4465)
rgb_std = (0.2023, 0.1994, 0.2010)
raw_vgg16 = vgg.vgg16(pretrained=True)
class MyRRN(nn.Module):
def __init__(self):
super(MyRRN, self).__init__()
self.features = raw_vgg16.features
# self.roi_pool = ROIPool((7, 7), 1/16)
self.deconv1 = nn.ConvTranspose2d(in_channels=512,
out_channels=64,
kernel_size=4,
stride=8,
padding=1)
self.conv1 = nn.Conv2d(in_channels=64,
out_channels=1,
kernel_size=3,
def main_worker(gpu, ngpus_per_node, args):
global best_acc1
args.gpu = gpu
if args.gpu is not None:
print("Use GPU: {} for training".format(args.gpu))
if args.distributed:
if args.dist_url == "env://" and args.rank == -1:
args.rank = int(os.environ["RANK"])
if args.multiprocessing_distributed:
# For multiprocessing distributed training, rank needs to be the
# global rank among all the processes
args.rank = args.rank * ngpus_per_node + gpu
dist.init_process_group(backend=args.dist_backend,
init_method=args.dist_url,
world_size=args.world_size,
rank=args.rank)
# create model
if args.pretrained:
print("=> using pre-trained model '{}'".format(args.arch))
model = models.__dict__[args.arch](pretrained=True)
else:
print("=> creating model '{}'".format(args.arch))
model = models.__dict__[args.arch]()
if args.arch == "vgg16":
model = vgg16()
if args.distributed:
# For multiprocessing distributed, DistributedDataParallel constructor
# should always set the single device scope, otherwise,
# DistributedDataParallel will use all available devices.
if args.gpu is not None:
torch.cuda.set_device(args.gpu)
model.cuda(args.gpu)
# When using a single GPU per process and per
# DistributedDataParallel, we need to divide the batch size
# ourselves based on the total number of GPUs we have
args.batch_size = int(args.batch_size / ngpus_per_node)
model = torch.nn.parallel.DistributedDataParallel(
model, device_ids=[args.gpu])
else:
model.cuda()
# DistributedDataParallel will divide and allocate batch_size to all
# available GPUs if device_ids are not set
model = torch.nn.parallel.DistributedDataParallel(model)
elif args.gpu is not None:
torch.cuda.set_device(args.gpu)
model = model.cuda(args.gpu)
else:
# DataParallel will divide and allocate batch_size to all available GPUs
if args.arch.startswith('alexnet') or args.arch.startswith('vgg'):
model.features = torch.nn.DataParallel(model.features)
model.cuda()
else:
model = torch.nn.DataParallel(model).cuda()
# define loss function (criterion) and optimizer
criterion = nn.CrossEntropyLoss().cuda(args.gpu)
optimizer = torch.optim.SGD(model.parameters(),
args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
best_acc1 = checkpoint['best_acc1']
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
print("=> loaded checkpoint '{}' (epoch {})".format(
args.resume, checkpoint['epoch']))
else:
print("=> no checkpoint found at '{}'".format(args.resume))
cudnn.benchmark = True
# Data loading code
traindir = os.path.join(args.data, 'train')
valdir = os.path.join(args.data, 'val')
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_dataset = datasets.ImageFolder(
traindir,
transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
]))
if args.distributed:
train_sampler = torch.utils.data.distributed.DistributedSampler(
train_dataset)
else:
train_sampler = None
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=args.batch_size,
shuffle=(train_sampler is None),
num_workers=args.workers,
pin_memory=True,
sampler=train_sampler)
val_loader = torch.utils.data.DataLoader(datasets.ImageFolder(
valdir,
transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])),
batch_size=args.batch_size,
shuffle=False,
num_workers=args.workers,
pin_memory=True)
if args.evaluate:
validate(val_loader, model, criterion, args)
return
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
train_sampler.set_epoch(epoch)
adjust_learning_rate(optimizer, epoch, args)
# train for one epoch
train(train_loader, model, criterion, optimizer, epoch, args)
# evaluate on validation set
acc1 = validate(val_loader, model, criterion, args)
# remember best acc@1 and save checkpoint
is_best = acc1 > best_acc1
best_acc1 = max(acc1, best_acc1)
if not args.multiprocessing_distributed or (
args.multiprocessing_distributed
and args.rank % ngpus_per_node == 0):
save_checkpoint(
{
'epoch': epoch + 1,
'arch': args.arch,
'state_dict': model.state_dict(),
'best_acc1': best_acc1,
'optimizer': optimizer.state_dict(),
}, is_best)
indices = range(start_point, end_point)
val_loader = torch.utils.data.DataLoader(
val_set,
batch_size=batch_size, shuffle=False,
num_workers=workers, pin_memory=True,
sampler=module.rangeSampler(indices))
#========================data loading=========================
#========================create model=========================
if model_name == 'vgg':
# quantization_factor = [(feature_bitwidth[j], weight_bitwidth[j]) for j in range(16)]
quantization_factor = [(feature_bitwidth[j], None) for j in range(16)]
model = vgg.vgg16(pretrained = True, bit_width = quantization_factor)
elif model_name == 'squeeze':
quantization_factor = [(feature_bitwidth[j], weight_bitwidth[j]) for j in range(26)]
model = squeeze.squeezenet1_0(pretrained = True, bit_width = quantization_factor)
model = torch.nn.DataParallel(model).cuda()
#========================create model=========================
#========================weight processing=========================
state_dict = model.state_dict()
counter = 0
for module_name, layer in model.named_modules():
if type(layer) == torch.nn.modules.conv.Conv2d \
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=args.batch_size,
shuffle=True,
pin_memory=True)
valid_dataset = datasets.ImageFolder(valid_path, transform=valid_transform)
valid_loader = torch.utils.data.DataLoader(valid_dataset,
batch_size=args.batch_size,
shuffle=False,
pin_memory=True)
# load prune model
checkpoint = torch.load(args.prune_path)
model = vgg16(pretrained=True).to(args.device)
mask = checkpoint['mask']
# Conv 5-3 [output]
model.features[-3] = conv_post_mask(model.features[-3], mask[0])
# FC 6 [input, output]
model.classifier[0] = linear_mask(model.classifier[0], mask[0], mask[1])
# FC 7 [input]
model.classifier[3] = linear_pre_mask(model.classifier[3], mask[1])
model.load_state_dict(checkpoint['state_dict'])
criterion = nn.CrossEntropyLoss().cuda()
optimizer = torch.optim.SGD(model.parameters(), args.lr, weight_decay=1e-4)
def showmodel():
network = vgg16(num_classes=10)
network = network.cuda()
summary(network, (3, 224, 224))
