def caffenet_multilabel(data_layer_params, datalayer):
# setup the python data layer
n = caffe.NetSpec()
n.data, n.label = L.Python(module='pascal_multilabel_datalayers',
layer=datalayer,
ntop=2,
param_str=str(data_layer_params))
# the net itself
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2)
n.pool5 = max_pool(n.relu5, 3, stride=2)
n.fc6, n.relu6 = fc_relu(n.pool5, 4096)
n.drop6 = L.Dropout(n.relu6, in_place=True)
n.fc7, n.relu7 = fc_relu(n.drop6, 4096)
n.drop7 = L.Dropout(n.relu7, in_place=True)
n.score = L.InnerProduct(n.drop7, num_output=20)
n.loss = L.SigmoidCrossEntropyLoss(n.score, n.label)
return str(n.to_proto())
def acn(source, loss_param, batch_size, include, shuffle, mirror):
n = caffe.NetSpec()
n.data, n.label = L.MultiImageData(include=include,
transform_param=dict(mirror=mirror, crop_size=227, mean_value=[104, 117, 123]),
multiimage_data_param=dict(root_folder="./../../data/RAP_dataset/", shuffle=shuffle, batch_size=batch_size, source=source, new_height=256, new_width=256),
ntop=2)
n.conv1 = L.Convolution(n.data, kernel_size=11, num_output=96, stride=4, pad=0, group=1,
param=[dict(name='conv1_w', lr_mult=1, decay_mult=1),
dict(name='conv1_b', lr_mult=2, decay_mult=0)])
n.relu1 = L.ReLU(n.conv1, in_place=True)
n.pool1 = L.Pooling(n.relu1, pool=P.Pooling.MAX, kernel_size=3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2 = L.Convolution(n.norm1, kernel_size=5, num_output=256, stride=1, pad=2, group=2,
param=[dict(name='conv2_w', lr_mult=1, decay_mult=1),
dict(name='conv2_b', lr_mult=2, decay_mult=0)])
n.relu2 = L.ReLU(n.conv2, in_place=True)
n.pool2 = L.Pooling(n.relu2, pool=P.Pooling.MAX, kernel_size=3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3 = L.Convolution(n.norm2, kernel_size=3, num_output=384, stride=1, pad=1, group=1,
param=[dict(name='conv3_w', lr_mult=1, decay_mult=1),
dict(name='conv3_b', lr_mult=2, decay_mult=0)])
n.relu3 = L.ReLU(n.conv3, in_place=True)
n.conv4 = L.Convolution(n.relu3, kernel_size=3, num_output=384, stride=1, pad=1, group=2,
param=[dict(name='conv4_w', lr_mult=1, decay_mult=1),
dict(name='conv4_b', lr_mult=2, decay_mult=0)])
n.relu4 = L.ReLU(n.conv4, in_place=True)
n.conv5 = L.Convolution(n.relu4, kernel_size=3, num_output=256, stride=1, pad=1, group=2,
param=[dict(name='conv5_w', lr_mult=1, decay_mult=1),
dict(name='conv5_b', lr_mult=2, decay_mult=0)])
n.relu5 = L.ReLU(n.conv5, in_place=True)
n.pool5 = L.Pooling(n.relu5, pool=P.Pooling.MAX, kernel_size=3, stride=2)
n.fc6 = L.InnerProduct(n.pool5, num_output=4096,
param=[dict(name='fc6_w', lr_mult=1, decay_mult=1),
dict(name='fc6_b', lr_mult=2, decay_mult=0)],
weight_filler=dict(type='gaussian', std=0.01), bias_filler=dict(type='constant', value=0))
n.relu6 = L.ReLU(n.fc6, in_place=True)
n.drop6 = L.Dropout(n.relu6, in_place=True)
n.fc7 = L.InnerProduct(n.drop6, num_output=4096,
param=[dict(name='fc7_w', lr_mult=1, decay_mult=1),
dict(name='fc7_b', lr_mult=2, decay_mult=0)],
weight_filler=dict(type='gaussian', std=0.01), bias_filler=dict(type='constant', value=0))
n.relu7 = L.ReLU(n.fc7, in_place=True)
n.drop7 = L.Dropout(n.relu7, in_place=True)
n.fchatt = L.InnerProduct(n.drop7, num_output=64,
param=[dict(name='fcatt_w', lr_mult=1, decay_mult=1),
dict(name='fcatt_b', lr_mult=2, decay_mult=0)],
weight_filler=dict(type='gaussian', std=0.01), bias_filler=dict(type='constant', value=0))
n.fine_fc8 = L.InnerProduct(n.fchatt, num_output=1,
param=[dict(name='fc8_body_w', lr_mult=1, decay_mult=1),
dict(name='fc8_body_b', lr_mult=2, decay_mult=0)],
weight_filler=dict(type='gaussian', std=0.01), bias_filler=dict(type='constant', value=0))
n.multilabel_loss = L.MultiLabelLoss(n.fine_fc8, n.label, loss_weight=1.0)
n.multilabel_accuracy = L.MultiLabelAccuracy(n.fine_fc8, n.label)
return n.to_proto()
def caffenet(n, nclasses, acclayer=False, softmax=False, learn=True):
# first input 'n', must have layer member: n.data
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4, learn=learn)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2, learn=learn)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1, learn=learn)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2, learn=learn)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2, learn=learn)
n.pool5 = max_pool(n.relu5, 3, stride=2)
n.fc6, n.relu6 = fc_relu(n.pool5, 4096, learn=learn)
n.drop6 = L.Dropout(n.relu6, in_place=True)
n.fc7, n.relu7 = fc_relu(n.drop6, 4096, learn=learn)
n.drop7 = L.Dropout(n.relu7, in_place=True)
n.score = L.InnerProduct(
n.drop7,
num_output=nclasses,
param=[dict(lr_mult=5, decay_mult=1),
dict(lr_mult=10, decay_mult=0)])
if softmax:
n.loss = L.SoftmaxWithLoss(n.score, n.label)
if acclayer:
n.acc = L.Accuracy(n.score, n.label)
return n
def mknet(lmdb, mean_file, batch_size, gg_max):
n = caffe.NetSpec()
n.data, n.label = L.Data(batch_size=batch_size, backend=P.Data.LMDB, source=lmdb,
transform_param=dict(mean_file=mean_file, scale=1./128), ntop=2)
n.conv1 = L.Convolution(n.data, kernel_size=5, num_output=32, pad=2, stride=1,
weight_filler=dict(type='gaussian', std=gg_max),
bias_filler=dict(type='constant'))
n.relu1 = L.ReLU(n.conv1, in_place=True)
n.pool1 = L.Pooling(n.relu1, kernel_size=3, stride=2, pool=P.Pooling.MAX)
n.norm1 = L.LRN(n.pool1, local_size=3, alpha=5e-5, beta=0.75, norm_region=1)
n.conv2 = L.Convolution(n.norm1, kernel_size=5, num_output=32, pad=2, stride=1,
weight_filler=dict(type='gaussian', std=gg_max),
bias_filler=dict(type='constant'))
n.relu2 = L.ReLU(n.conv2, in_place=True)
n.pool2 = L.Pooling(n.relu2, kernel_size=3, stride=2, pool=P.Pooling.AVE)
n.norm2 = L.LRN(n.pool2, local_size=3, alpha=5e-5, beta=0.75, norm_region=1)
n.conv3 = L.Convolution(n.norm2, kernel_size=5, num_output=64, pad=2, stride=1,
weight_filler=dict(type='gaussian', std=gg_max),
bias_filler=dict(type='constant'))
n.relu3 = L.ReLU(n.conv3, in_place=True)
n.pool3 = L.Pooling(n.relu3, kernel_size=3, stride=2, pool=P.Pooling.AVE)
n.fc1 = L.InnerProduct(n.pool3, num_output=10,
weight_filler=dict(type='gaussian', std=gg_max),
bias_filler=dict(type='constant'))
n.loss = L.SoftmaxWithLoss(n.fc1, n.label)
n.acc = L.Accuracy(n.fc1, n.label)
return n.to_proto()
def caffenet(batch_size=256, include_acc=False, train = True, learn_all=True):
subset = 'train' if train else 'test'
source = caffe_root + 'data/flickr_style/%s.txt' % subset
transform_param = dict(mirror=train, crop_size=227,
mean_file=caffe_root + 'data/ilsvrc12/imagenet_mean.binaryproto')
n=caffe.NetSpec()
n.style_data, n.style_label = L.ImageData(
transform_param=transform_param, source=source,
batch_size=50, new_height=256, new_width=256, ntop=2)
param = learned_param if learn_all else frozen_param
# the net itself
n.conv1, n.relu1 = conv_relu(n.style_data, 11, 96, stride=4, param=param)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2, param=param)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1, param=param)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2, param=param)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2, param=param)
n.pool5 = max_pool(n.relu5, 3, stride=2)
n.fc6, n.relu6 = fc_relu(n.pool5, 4096, param=param)
n.drop6 = L.Dropout(n.relu6, in_place=True)
n.fc7, n.relu7 = fc_relu(n.drop6, 4096, param=param)
n.drop7 = L.Dropout(n.relu7, in_place=True)
n.fc8_style = L.InnerProduct(n.drop7, num_output=CLASS_NUM, param=learned_param)
n.loss = L.SoftmaxWithLoss(n.fc8_style, n.style_label)
n.acc = L.Accuracy(n.fc8_style, n.style_label)
return n.to_proto()
def caffenet(lmdb, batch_size=256, include_acc=False):
data, label = L.Data(source=lmdb, backend=P.Data.LMDB, batch_size=batch_size, ntop=2,
transform_param=dict(crop_size=227, mean_value=[104, 117, 123], mirror=True))
# the net itself
conv1, relu1 = conv_relu(data, 11, 96, stride=4)
pool1 = max_pool(relu1, 3, stride=2)
norm1 = L.LRN(pool1, local_size=5, alpha=1e-4, beta=0.75)
conv2, relu2 = conv_relu(norm1, 5, 256, pad=2, group=2)
pool2 = max_pool(relu2, 3, stride=2)
norm2 = L.LRN(pool2, local_size=5, alpha=1e-4, beta=0.75)
conv3, relu3 = conv_relu(norm2, 3, 384, pad=1)
conv4, relu4 = conv_relu(relu3, 3, 384, pad=1, group=2)
conv5, relu5 = conv_relu(relu4, 3, 256, pad=1, group=2)
pool5 = max_pool(relu5, 3, stride=2)
fc6, relu6 = fc_relu(pool5, 4096)
drop6 = L.Dropout(relu6, in_place=True)
fc7, relu7 = fc_relu(drop6, 4096)
drop7 = L.Dropout(relu7, in_place=True)
fc8 = L.InnerProduct(drop7, num_output=1000)
loss = L.SoftmaxWithLoss(fc8, label)
if include_acc:
acc = L.Accuracy(fc8, label)
return to_proto(loss, acc)
else:
return to_proto(loss)
def caffenet(data,
label=None,
train=True,
num_classes=1000,
classifier_name='fc8',
learn_all=False):
"""Returns a NetSpec specifying CaffeNet, following the original proto text
specification (./models/bvlc_reference_caffenet/train_val.prototxt)."""
n = caffe.NetSpec()
n.data = data
# frozen ?? #
# 解释：如果只训练最后一层即全连接层，那么其他层的参数都不变也就是frozen
param = learned_param if learn_all else frozen_param
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4, param=param)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2, param=param)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1, param=param)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2, param=param)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2, param=param)
n.pool5 = max_pool(n.relu5, 3, stride=2)
n.fc6, n.relu6 = fc_relu(n.pool5, 4096, param=param)
## 这个if else 目的是什么？ ##
# 如果只预测不训练，params取多少都无所谓，因为不用反向传播，如果需要训练，那么就得分情况啦
# 如果训练，则还要增加一个层：L.Dropout
if train:
n.drop6 = fc7input = L.Dropout(n.relu6, in_place=True)
else:
fc7input = n.relu6
n.fc7, n.relu7 = fc_relu(fc7input, 4096, param=param)
if train:
n.drop7 = fc8input = L.Dropout(n.relu7, in_place=True)
else:
fc8input = n.relu7
# always learn fc8 (param=learned_param)
fc8 = L.InnerProduct(fc8input, num_output=num_classes, param=learned_param)
# give fc8 the name specified by argument `classifier_name`
n.__setattr__(classifier_name, fc8)
# 对于只预测不训练，当然只是输出每种类型对应的可能结果啦
if not train:
n.probs = L.Softmax(fc8)
# 对于要训练的，那么就是得输出训练集的损失函数和正确率啦
if label is not None:
n.label = label
n.loss = L.SoftmaxWithLoss(fc8, n.label)
n.acc = L.Accuracy(fc8, n.label)
# write the net to a temporary file and return its filename
with tempfile.NamedTemporaryFile(delete=False) as f:
f.write(str(n.to_proto()))
print(n.to_proto())
return f.name
def generate_net():
n = caffe.NetSpec()
n.data = L.Data(source=TRAIN_LMDB_DATA_FILE,
backend=P.Data.LMDB,
batch_size=1,
ntop=1,
transform_param=dict(scale=1. / 255))
n.label = L.Data(source=TRAIN_LMDB_LABEL_FILE,
backend=P.Data.LMDB,
batch_size=1,
ntop=1)
# the base net
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4, pad=100)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2)
n.pool5 = max_pool(n.relu5, 3, stride=2)
# fully conv
n.fc6, n.relu6 = conv_relu(n.pool5, 6, 4096)
n.drop6 = L.Dropout(n.relu6, dropout_ratio=0.5, in_place=True)
n.fc7, n.relu7 = conv_relu(n.drop6, 1, 4096)
n.drop7 = L.Dropout(n.relu7, dropout_ratio=0.5, in_place=True)
# weight_filler=dict(type='gaussian', std=0.0001), bias_filler=dict(type='constant')
n.score_fr_ = L.Convolution(
n.drop7,
num_output=2,
kernel_size=1,
pad=0,
param=[dict(lr_mult=1, decay_mult=1),
dict(lr_mult=2, decay_mult=0)])
n.upscore_ = L.Deconvolution(n.score_fr_,
convolution_param=dict(
num_output=2,
kernel_size=63,
stride=32,
group=2,
bias_term=False,
weight_filler=dict(type='bilinear')),
param=[dict(lr_mult=0)])
n.score = crop(n.upscore_, n.data)
n.loss = L.SoftmaxWithLoss(n.score,
n.label,
loss_weight=1,
loss_param=dict(normalize=False,
ignore_label=255))
return n.to_proto()
def fcn(split):
n = caffe.NetSpec()
pydata_params = dict(split=split,
mean=(104.00699, 116.66877, 122.67892),
seed=1337)
if split == 'train':
pydata_params['sbdd_dir'] = '../data/sbdd/dataset'
pylayer = 'BDDSegDataLayer'
else:
pydata_params['bdd_dir'] = '/dl/data/bdd100k/seg'
pylayer = 'BDDSegDataLayer'
n.data, n.label = L.Python(module='voc_layers',
layer=pylayer,
ntop=2,
param_str=str(pydata_params))
# the base net
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4, pad=100)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2)
n.pool5 = max_pool(n.relu5, 3, stride=2)
# fully conv
n.fc6, n.relu6 = conv_relu(n.pool5, 6, 4096)
n.drop6 = L.Dropout(n.relu6, dropout_ratio=0.5, in_place=True)
n.fc7, n.relu7 = conv_relu(n.drop6, 1, 4096)
n.drop7 = L.Dropout(n.relu7, dropout_ratio=0.5, in_place=True)
n.score_fr = L.Convolution(
n.drop7,
num_output=21,
kernel_size=1,
pad=0,
param=[dict(lr_mult=1, decay_mult=1),
dict(lr_mult=2, decay_mult=0)])
# upscore
n.upscore = L.Deconvolution(n.score_fr,
convolution_param=dict(num_output=21,
kernel_size=63,
stride=32,
bias_term=False),
param=[dict(lr_mult=0)])
n.score = crop(n.upscore, n.data)
n.loss = L.SoftmaxWithLoss(n.score,
n.label,
loss_param=dict(normalize=True,
ignore_label=255))
return n.to_proto()
def caffenet_body(net, data, post, is_train):
# the net itself
net['conv1' + post], net['relu1' + post] = conv_relu(net[data],
11,
96,
stride=4,
is_train=is_train)
net['pool1' + post] = max_pool(net['relu1' + post], 3, stride=2)
net['norm1' + post] = L.LRN(net['pool1' + post],
local_size=5,
alpha=1e-4,
beta=0.75,
engine=P.LRN.CAFFE)
net['conv2' + post], net['relu2' + post] = conv_relu(net['norm1' + post],
5,
256,
pad=2,
group=2,
is_train=is_train)
net['pool2' + post] = max_pool(net['relu2' + post], 3, stride=2)
net['norm2' + post] = L.LRN(net['pool2' + post],
local_size=5,
alpha=1e-4,
beta=0.75,
engine=P.LRN.CAFFE)
net['conv3' + post], net['relu3' + post] = conv_relu(net['norm2' + post],
3,
384,
pad=1,
is_train=is_train)
net['conv4' + post], net['relu4' + post] = conv_relu(net['relu3' + post],
3,
384,
pad=1,
group=2,
is_train=is_train)
net['conv5' + post], net['relu5' + post] = conv_relu(net['relu4' + post],
3,
256,
pad=1,
group=2,
is_train=is_train)
net['pool5' + post] = max_pool(net['relu5' + post], 3, stride=2)
net['fc6' + post], net['relu6' + post] = fc_relu(net['pool5' + post],
4096,
is_train=is_train)
net['drop6' + post] = L.Dropout(net['relu6' + post], in_place=True)
net['fc7' + post], net['relu7' + post] = fc_relu(net['drop6' + post],
4096,
is_train=is_train)
net['drop7' + post] = L.Dropout(net['relu7' + post], in_place=True)
#n.score = L.InnerProduct(n.drop7, num_output=20, weight_filler=dict(type='gaussian', std=0.01))
#n.loss = L.SigmoidCrossEntropyLoss(n.score, n.label)
final = 'drop7' + post
return net, final
def cnn(split):
n = caffe.NetSpec()
pydata_params = dict(dataset_dir='/home/kevin/dataset/processed_data', variable='depth_map', split=split, mean=(2),
seed=1337, batch_size=1, img_size=(250,250))
if split == 'deploy':
n.img = L.Input(name='input', ntop=2, shape=[dict(dim=1),dict(dim=1),dict(dim=250),dict(dim=250)])
else:
if split is 'train':
pydata_params['dtype'] = 'object'
pylayer = 'ModelNetDataLayer'
else:
pydata_params['dtype'] = 'object'
pylayer = 'ModelNetDataLayer'
n.img, n.label = L.Python(module='data_layers.model_net_layer', layer=pylayer,
ntop=2, param_str=str(pydata_params))
# the base net
n.conv1, n.relu1 = conv_relu("conv1", n.img, 96, ks=11, stride=4, pad=0)
n.pool1 = max_pool(n.relu1, ks=3)
n.norm1 = L.LRN(n.pool1, lrn_param=dict(local_size=5, alpha=0.0005, beta=0.75, k=2))
# n.bn1 = L.BatchNorm(n.pool1, param=[dict(lr_mult=0),dict(lr_mult=0),dict(lr_mult=0)], batch_norm_param=dict(use_global_stats=True))
n.conv2, n.relu2 = conv_relu("conv2", n.norm1, 256, ks=5, pad=2, group=2)
n.pool2 = max_pool(n.relu2, ks=3)
n.norm2 = L.LRN(n.pool2, lrn_param=dict(local_size=5, alpha=0.0005, beta=0.75, k=2))
# n.bn2 = L.BatchNorm(n.pool2, param=[dict(lr_mult=0),dict(lr_mult=0),dict(lr_mult=0)], batch_norm_param=dict(use_global_stats=True))
n.conv3, n.relu3 = conv_relu("conv3", n.norm2, 384, ks=3, pad=1, group=2)
n.conv4, n.relu4 = conv_relu("conv4", n.relu3, 256, ks=3, pad=1, group=2)
n.pool5 = max_pool(n.relu4, ks=3)
n.fc6, n.relu6 = fc_relu(n.pool5, 4096, lr1=1, lr2=2)
n.drop6 = L.Dropout(n.relu6, dropout_ratio=0.5, in_place=True)
n.fc7, n.relu7 = fc_relu(n.drop6 , 4096, lr1=1, lr2=2)
n.drop7 = L.Dropout(n.relu7, dropout_ratio=0.5, in_place=True)
n.fc8 = fc(n.drop7, 40, lr1=1, lr2=2)
if split != 'deploy':
n.accuracy = L.Accuracy(n.fc8, n.label)
#n.loss = L.SoftmaxWithLoss(n.fc8, n.label)
n.loss = L.Python(n.fc8, n.label, loss_weight=1, module='nn_layers.max_softmax_loss_layer', layer='MaxSoftmaxLossLayer')
# n.display = L.Scale(n.corr, param=[dict(lr_mult=0)], filler=dict(type='constant',value=1.0))
# n.fc9_bn = L.BatchNorm(n.relu9, param=[dict(lr_mult=0),dict(lr_mult=0),dict(lr_mult=0)], batch_norm_param=dict(use_global_stats=True))
return n.to_proto()
def ZFNetBody(net, from_layer, for_training=True):
net.conv1 = L.Convolution(net[from_layer], kernel_size=k_conv1, stride=s_conv1, num_output=d_conv1, pad=p_conv1,
bias_term=True, weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
net.relu1 = L.ReLU(net.conv1, in_place=True)
net.pool1 = L.Pooling(net.relu1, pool=P.Pooling.MAX, kernel_size=k_pool1, stride=s_pool1)
net.norm1 = L.LRN(net.pool1, lrn_param=dict(local_size=local_size_norm1, alpha=alpha_norm1, beta=beta_norm1))
net.conv2 = L.Convolution(net.norm1, kernel_size=k_conv2, stride=s_conv2, num_output=d_conv2, #pad=p_conv2,
bias_term=True, weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
net.relu2 = L.ReLU(net.conv2, in_place=True)
net.pool2 = L.Pooling(net.relu2, pool=P.Pooling.MAX, kernel_size=k_pool2, stride=s_pool2)
net.norm2 = L.LRN(net.pool2, lrn_param=dict(local_size=local_size_norm2, alpha=alpha_norm2, beta=beta_norm2))
net.conv3 = L.Convolution(net.norm2, kernel_size=k_conv3, stride=s_conv3, num_output=d_conv3, pad=p_conv3,
bias_term=True, weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
net.relu3 = L.ReLU(net.conv3, in_place=True)
net.conv4 = L.Convolution(net.relu3, kernel_size=k_conv4, stride=s_conv4, num_output=d_conv4, pad=p_conv4,
bias_term=True, weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
net.relu4 = L.ReLU(net.conv4, in_place=True)
net.conv5 = L.Convolution(net.relu4, kernel_size=k_conv5, stride=s_conv5, num_output=d_conv5, pad=p_conv5,
bias_term=True, weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
net.relu5 = L.ReLU(net.conv5, in_place=True)
net.pool5 = L.Pooling(net.relu5, pool=P.Pooling.MAX, kernel_size=k_pool5, stride=s_pool5)
net.fc6 = L.InnerProduct(net.pool5, num_output=k_ip6,
weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
net.relu6 = L.ReLU(net.fc6, in_place=True)
net.drop6 = L.Dropout(net.relu6, dropout_param=dict(dropout_ratio=r_drop6), in_place=True)
net.fc7 = L.InnerProduct(net.fc6, num_output=k_ip7,
weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
net.relu7 = L.ReLU(net.fc7, in_place=True)
net.drop7 = L.Dropout(net.relu7, dropout_param=dict(dropout_ratio=r_drop7), in_place=True)
net.fc8 = L.InnerProduct(net.fc7, num_output=k_ip8,
weight_filler=dict(type='gaussian',std=0.01), bias_filler=dict(type='constant',std=0),
param=[dict(lr_mult=1,decay_mult=1), dict(lr_mult=2,decay_mult=0)])
if not for_training:
net.acc = L.Accuracy(net.fc8, net.label, include=dict(phase=caffe_pb2.Phase.Value('TEST')))
net.acc5 = L.Accuracy(net.fc8, net.label, include=dict(phase=caffe_pb2.Phase.Value('TEST')), accuracy_param=dict(top_k=5))
net.loss = L.SoftmaxWithLoss(net.fc8, net.label)
return net
def caffenet(data,
label=None,
train=True,
num_classes=1000,
classifier_name='fc8',
learn_all=False,
deploy=False):
"""Returns a NetSpec specifying CaffeNet, following the original proto text
specification (./models/bvlc_reference_caffenet/train_val.prototxt)."""
n = caffe.NetSpec()
n.data = data
param = learned_param if learn_all else frozen_param
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4, param=param)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2, param=param)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1, param=param)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2, param=param)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2, param=param)
n.pool5 = max_pool(n.relu5, 3, stride=2)
n.fc6, n.relu6 = fc_relu(n.pool5, 4096, param=param)
if train:
n.drop6 = fc7input = L.Dropout(n.relu6, in_place=True)
else:
fc7input = n.relu6
n.fc7, n.relu7 = fc_relu(fc7input, 4096, param=param)
if train:
n.drop7 = fc8input = L.Dropout(n.relu7, in_place=True)
else:
fc8input = n.relu7
# always learn fc8 (param=learned_param)
fc8 = L.InnerProduct(fc8input, num_output=num_classes, param=learned_param)
# give fc8 the name specified by argument `classifier_name`
n.__setattr__(classifier_name, fc8)
if not train:
n.probs = L.Softmax(fc8)
if label is not None:
n.label = label
n.loss = L.SoftmaxWithLoss(fc8, n.label)
n.acc = L.Accuracy(fc8, n.label)
if train:
protxt_file = train_prototxt
else:
protxt_file = val_prototxt
if deploy:
protxt_file = deploy_style_recognition_net_filename
with open(protxt_file, 'w') as f:
f.write(str(n.to_proto()))
def define_network(args, imageFile, vidIds, radarFiles, training=False):
net = caffe.NetSpec()
# Setting up data layer
transformParam = dict(mirror=training, mean_value = args.mean)
pydataParams = dict(radar_files = radarFiles, videos = vidIds, batch_size = args.batchSize)
net.data, net.label = L.ImageData(transform_param = transformParam, source=imageFile, shuffle=False, batch_size=args.batchSize, ntop=2)
if args.expType != 'image':
net.radar = L.Python(module='radarDataLayer', layer='RadarDataLayer', param_str=str(pydataParams), ntop=1)
if args.expType == "joint" or args.expType == "image":
net.conv1, net.relu1 = conv_relu(net.data, 11, 96, stride=4)
net.pool1 = max_pool(net.relu1, 3, stride=2)
net.norm1 = L.LRN(net.pool1, local_size=5, alpha=1e-4, beta=0.75)
net.conv2, net.relu2 = conv_relu(net.norm1, 5, 256, pad=2, group=2)
net.pool2 = max_pool(net.relu2, 3, stride=2)
net.norm2 = L.LRN(net.pool2, local_size=5, alpha=1e-4, beta=0.75)
net.conv3, net.relu3 = conv_relu(net.norm2, 3, 384, pad=1)
net.conv4, net.relu4 = conv_relu(net.relu3, 3, 384, pad=1, group=2)
net.conv5, net.relu5 = conv_relu(net.relu4, 3, 256, pad=1, group=2)
net.pool5 = max_pool(net.relu5, 3, stride=2)
net.fc6_new, net.relu6_new = fc_relu(net.pool5, 4096)
net.drop6 = L.Dropout(net.relu6_new, in_place=True)
net.fc7_new = L.InnerProduct(net.drop6, num_output=4096, param=learned_param, weight_filler=fc_filler)
if args.expType == "joint":
net.concat = L.Concat(net.fc7_new, net.radar)
net.relu7 = L.ReLU(net.concat, in_place=True)
else:
net.relu7 = L.ReLU(net.fc7_new, in_place=True)
net.drop7 = L.Dropout(net.relu7, in_place=True)
net.final = L.InnerProduct(net.drop7, num_output=args.num_out, param=learned_param, weight_filler=fc_filler)
elif args.expType == "radar":
net.silence = L.Silence(net.data, ntop=0)
net.fc7_new = L.InnerProduct(net.radar, num_output=1024, param=learned_param, weight_filler=fc_filler)
net.relu7 = L.ReLU(net.fc7_new, in_place=True)
net.drop7 = L.Dropout(net.relu7, in_place=True)
net.final = L.InnerProduct(net.drop7, num_output=args.num_out, param=learned_param, weight_filler=fc_filler)
net.loss = L.SoftmaxWithLoss(net.final, net.label)
net.acc = L.Accuracy(net.final, net.label)
return net.to_proto()
def conv1_to_5(n, param):
n.conv1, n.relu1 = conv_relu(n.data, ks=7, nout=96, stride=2, pad=3, param=param)
n.norm1 = L.LRN(n.conv1, local_size=3, alpha=0.00005, beta=0.75, norm_region=LRN_NORMREGION_WITHIN_CHANNEL,
engine=LRN_ENGINE_CAFFE)
n.pool1 = max_pool(n.norm1, 3, stride=2)
n.conv2, n.relu2 = conv_relu(n.pool1, 5, 256, stride=2, pad=2, param=param)
n.norm2 = L.LRN(n.conv2, local_size=3, alpha=0.00005, beta=0.75, norm_region=LRN_NORMREGION_WITHIN_CHANNEL,
engine=LRN_ENGINE_CAFFE)
n.pool2 = max_pool(n.norm2, ks=3, stride=2, pad=1)
n.conv3, n.relu3 = conv_relu(n.pool2, ks=3, nout=384, stride=1, pad=1, param=param)
n.conv4, n.relu4 = conv_relu(n.conv3, ks=3, nout=384, stride=1, pad=1, param=param)
n.conv5, n.relu5 = conv_relu(n.conv4, ks=3, nout=256, stride=1, pad=1, param=param)
def conv_bn_layers(bottom, num_filter, param, weight_filler, bias_filler, kernel = 3, stride=1, pad = 0):
conv = conv_layer(bottom, num_filter = num_filter, kernel = kernel, stride=stride, pad = pad,
param = param, weight_filler = weight_filler, bias_filler = bias_filler)
#bn = L.BatchNorm(conv, in_place=True, param = [{"lr_mult":0},{"lr_mult":0},{"lr_mult":0}]) #use_global_stats = False,
bn = L.BN(conv, param = [dict(lr_mult=1), dict(lr_mult=1)], scale_filler=dict(type="constant", value=1), shift_filler=dict(type="constant", value=0))
lrn = L.LRN(bn)
return lrn
def assemble_net(self, lmdb, mean_file, proto_filename, batch_size, crop_size,img_channels, phase):
layer_dict = {}
n = caffe.NetSpec()
if phase in ['Train', 'TRAIN','train']:
dropout_flag = True
loss_layer_flag = True
prob_layer_flag = False
acc_layer_flag = True
if phase in ['test', 'TEST', 'Test']:
dropout_flag = False
loss_layer_flag = True
prob_layer_flag = False
acc_layer_flag = True
if phase in ['deploy', 'DEPLOY', 'Deploy']:
dropout_flag = False
loss_layer_flag = False
prob_layer_flag = True
acc_layer_flag = False
if phase in ['deploy', 'DEPLOY', 'Deploy']:
n.data = L.Input(input_param= {'shape':{'dim':[1,img_channels,crop_size,crop_size]}})
else:
n.data, n.label = L.Data(transform_param={'crop_size': crop_size, 'mean_file':mean_file}, data_param={'source':lmdb,'batch_size':batch_size, 'backend': P.Data.LMDB},
ntop=2)
for i, layer_name in enumerate(self.topology):
#print layer_name
if 'conv' in layer_name:
if i==0:
n.conv = self.conv2d(n.data, layer_name)
else:
n.conv = self.conv2d(layer_dict[str(i-1)], layer_name)
layer_dict[str(i)] = n.conv
elif 'pool' in layer_name:
pool_params = {'pool': 0, 'kernel_size': 3, 'stride': 2}
n.pool = L.Pooling(layer_dict[str(i - 1)], pooling_param=pool_params, name=layer_name)
layer_dict[str(i)] = n.pool
elif 'fc' in layer_name or 'output' in layer_name:
n.fc = self.fc(layer_dict[str(i - 1)], layer_name, dropout_flag)
layer_dict[str(i)] = n.fc
elif 'lrn' in layer_name:
n.norm = L.LRN(layer_dict[str(i - 1)], local_size=5, alpha=1e-4, beta=0.75, name=layer_name)
layer_dict[str(i)] = n.norm
if loss_layer_flag:
n.loss = L.SoftmaxWithLoss(layer_dict[str(i)], n.label, name='loss')
if prob_layer_flag:
n.prob = L.Softmax(layer_dict[str(i)], name='loss')
if acc_layer_flag:
n.accuracy = L.Accuracy(layer_dict[str(i)], n.label, name='accuracy')
print layer_dict
with open(proto_filename, 'w') as f:
f.write(str(n.to_proto()))
def convLayer(prev, lrn=False, param_name=None, bn=False, **kwargs):
if param_name:
name1 = param_name + '_kernels'
name2 = param_name + '_bias'
conv = L.Convolution(
prev,
param=[dict(lr_mult=1, name=name1),
dict(lr_mult=2, name=name2)],
weight_filler=dict(type='msra'),
**kwargs)
else:
conv = L.Convolution(prev,
param=[dict(lr_mult=1),
dict(lr_mult=2)],
weight_filler=dict(type='msra'),
**kwargs)
if bn:
bn = L.BatchNorm(conv)
relu = L.ReLU(bn, in_place=True)
else:
relu = L.ReLU(conv, in_place=True)
if lrn:
# optional Local Response Normalization
relu = L.LRN(relu,
lrn_param={
'local_size': min(kwargs['num_output'] / 3, 5),
'alpha': 0.0001,
'beta': 0.75
})
return relu
def lenet(lmdb, batch_size):
# our version of LeNet: a series of linear and simple nonlinear transformations
n = caffe.NetSpec()
n.data, n.label = L.Data(batch_size=batch_size,
backend=P.Data.LMDB,
source=lmdb,
transform_param=dict(scale=1. / 255),
ntop=2)
n.conv1 = L.Convolution(n.data,
kernel_size=5,
num_output=32,
weight_filler=dict(type='xavier'))
n.pool1 = L.Pooling(n.conv1, kernel_size=3, stride=2, pool=P.Pooling.MAX)
n.relu1 = L.ReLU(n.pool1)
n.norm1 = L.LRN(n.pool1, local_size=3, alpha=.00005, beta=.75)
n.conv2 = L.Convolution(n.norm1,
kernel_size=3,
num_output=64,
weight_filler=dict(type='xavier'))
n.pool2 = L.Pooling(n.conv2, kernel_size=2, stride=2, pool=P.Pooling.MAX)
n.conv3 = L.Convolution(n.pool1,
kernel_size=3,
num_output=128,
weight_filler=dict(type='xavier'))
n.pool3 = L.Pooling(n.conv2, kernel_size=2, stride=2, pool=P.Pooling.MAX)
n.ip1 = L.InnerProduct(n.pool2,
num_output=128 * 3 * 3,
weight_filler=dict(type='xavier'))
n.relu1 = L.ReLU(n.ip1, in_place=True)
n.ip2 = L.InnerProduct(n.relu1,
num_output=10,
weight_filler=dict(type='xavier'))
n.loss = L.SoftmaxWithLoss(n.ip2, n.label)
return n.to_proto()
def prototxt_rep(r):
n = caffe.NetSpec()
layers = r['model_state']['layers']
size = math.sqrt(layers['data']['outputs']/3.)
n.data, n.label = L.DummyData(shape=[dict(dim=[1, 3, size, size]),
dict(dim=[1, 1, 1, 1])])
lays['data'] = n.data
lays['label'] = n.label
for lname in layer_names:
layer0 = layers[lname]
input_name = layer0['inputs'][0]
if input_name + '_neuron' in lays:
input_name = input_name +'_neuron'
intput = lays[input_name]
ltype == layer0['type']
if ltype == 'conv':
lay = L.Convolution(input,
kernel_size = layer0['filterSize'][0],
pad = -layer0['padding'][0],
stride = layer0['stride'][0],
num_output = layer0['filters'])
elif ltype == 'pool':
ptype = P.Pooling.MAX if layer0['pool'] == 'max' else P.Pooling.AVE
lay = L.Pooling(input,
kernel_size = layer0['sizeX'],
stride = layer0['stride'],
pool = ptype)
elif ltype == 'fc':
lay = L.InnerProduct(input,
num_output = layer0['outputs'])
elif ltype == 'neuron':
ntype = layer0['neuron']
if ntype == 'relu':
cname = 'ReLU'
lay = getattr(L, cname)(input, in_place=True)
elif ltype == 'cmrnorm':
lay = L.LRN(input,
local_size = layer0['size'],
alpha = layer0['scale'],
beta = layer0['pow'])
elif ltype == 'dropout2':
lay = L.Dropout(input,
dropout_ratio = 0.5)
else:
raise ValueError("Layer type %s not recognized" % ltype)
setattr(n, lname, lay)
lays[lname] = lay
if layer0.get('neuron'):
ntype = layer0['neuron']
nname = lname + '_neuron'
if ntype == 'relu':
cname = 'ReLU'
nlay = getattr(L, cname)(lay, in_place=True)
setattr(n, nname, nlay)
lays[nname] = nlay
return n
def minivggnet(data,
labels=None,
train=False,
cudnn=False,
param=learned_param,
num_classes=100,
with_labels=True):
"""
Returns a protobuf text file specifying a variant of VGG
"""
n = caffe.NetSpec()
n.data = data
conv_kwargs = dict(param=param, train=train, cudnn=cudnn)
n.conv1, n.relu1 = conv_relu(n.data, 7, 96, stride=2, **conv_kwargs)
n.norm1 = L.LRN(n.relu1, local_size=5, alpha=0.0005, beta=0.75, k=2)
n.pool1 = max_pool(n.norm1, 3, stride=3, train=train, cudnn=cudnn)
n.conv2, n.relu2 = conv_relu(n.pool1,
5,
256,
pad=1,
stride=2,
group=2,
**conv_kwargs)
n.pool2 = max_pool(n.relu2, 2, stride=2, train=train, cudnn=cudnn)
n.conv3, n.relu3 = conv_relu(n.pool2, 3, 512, pad=1, **conv_kwargs)
n.conv4, n.relu4 = conv_relu(n.relu3,
3,
512,
pad=1,
group=2,
**conv_kwargs)
n.conv5, n.relu5 = conv_relu(n.relu4,
3,
512,
pad=1,
group=2,
**conv_kwargs)
n.pool5 = max_pool(n.relu5, 3, stride=3, train=train, cudnn=cudnn)
n.fc6, n.relu6 = fc_relu(n.pool5, 1024, param=param)
n.drop6 = L.Dropout(n.relu6, in_place=True)
n.fc7, n.relu7 = fc_relu(n.drop6, 1024, param=param)
n.drop7 = L.Dropout(n.relu7, in_place=True)
preds = n.fc8 = L.InnerProduct(n.drop7,
num_output=num_classes,
param=param)
if not train:
# Compute the per-label probabilities at test/inference time.
preds = n.probs = L.Softmax(n.fc8)
if with_labels:
n.label = labels
n.loss = L.SoftmaxWithLoss(n.fc8, n.label)
n.accuracy_at_1 = L.Accuracy(preds, n.label)
n.accuracy_at_5 = L.Accuracy(preds,
n.label,
accuracy_param=dict(top_k=5))
else:
n.ignored_label = labels
n.silence_label = L.Silence(n.ignored_label, ntop=0)
return to_tempfile(str(n.to_proto()))
def LRN(bottom, name='lrn', ls=5, a=0.0001, b=0.75):
return L.LRN(bottom,
name=name,
lrn_param={
'local_size': ls,
'alpha': a,
'beta': b
})
def net():
n = caffe.NetSpec()
n.data = L.Input(input_param=dict(shape=dict(dim=data_shape)))
n.dataout = L.LRN(n.data,
local_size=_local_size,
alpha=_alpha,
beta=_beta)
return n.to_proto()
def caffenet(data,
label=None,
train=True,
num_classes=1000,
classifier_name='fc8',
learn_all=False):
# Returns a NetSpec specifying CaffeNet
n = caffe.NetSpec()
n.data = data
param = learned_param if learn_all else frozen_param
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4, param=param)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2, param=param)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1, param=param)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2, param=param)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2, param=param)
n.pool5 = max_pool(n.relu5, 3, stride=2)
n.fc6, n.relu6 = fc_relu(n.pool5, 1024, param=param)
if train:
n.drop6 = fc7input = L.Dropout(n.relu6, in_place=True)
else:
fc7input = n.relu6
n.fc7, n.relu7 = fc_relu(fc7input, 1024, param=param)
if train:
n.drop7 = fc8input = L.Dropout(n.relu7, in_place=True)
else:
fc8input = n.relu7
# always learn fc8 (param=learned_param)
fc8 = L.InnerProduct(fc8input, num_output=num_classes, param=learned_param)
# give fc8 the name specified by argument `classifier_name`
n.__setattr__("fc6_m", n.fc6)
n.__setattr__("fc7_m", n.fc7)
n.__setattr__(classifier_name, fc8)
if not train:
n.probs = L.Softmax(fc8)
if label is not None:
n.label = label
n.loss = L.SoftmaxWithLoss(fc8, n.label)
n.acc = L.Accuracy(fc8, n.label)
# write the net to a temporary file and return its filename
with tempfile.NamedTemporaryFile(delete=False) as f:
f.write(str(n.to_proto()))
return f.name
def test_lrn():
# type: ()->caffe.NetSpec
n = caffe.NetSpec()
n.input1 = L.Input(shape=make_shape([10, 3, 64, 64]))
n.lrn1 = L.LRN(n.input1)
return n
def add_caffenet(net, num_labels):
net.conv1, net.relu1 = conv_relu(net.data, 11, 96, stride=4)
net.pool1 = max_pool(net.relu1, 3, stride=2)
net.norm1 = L.LRN(net.pool1, local_size=5, alpha=1e-4, beta=0.75)
net.conv2, net.relu2 = conv_relu(net.norm1, 5, 256, pad=2, group=2)
net.pool2 = max_pool(net.relu2, 3, stride=2)
net.norm2 = L.LRN(net.pool2, local_size=5, alpha=1e-4, beta=0.75)
net.conv3, net.relu3 = conv_relu(net.norm2, 3, 384, pad=1)
net.conv4, net.relu4 = conv_relu(net.relu3, 3, 384, pad=1, group=2)
net.conv5, net.relu5 = conv_relu(net.relu4, 3, 256, pad=1, group=2)
net.pool5 = max_pool(net.relu5, 3, stride=2)
net.fc6, net.relu6 = fc_relu(net.pool5, 4096)
net.drop6 = L.Dropout(net.relu6, in_place=True)
net.fc7, net.relu7 = fc_relu(net.drop6, 4096)
net.drop7 = L.Dropout(net.relu7, in_place=True)
net.score = L.InnerProduct(net.drop7, num_output=num_labels)
return net
def add_lrn(net, bottom, name, local_size, alpha, beta, k):
"""Add local response normalizaiton unit """
net[name] = L.LRN(bottom,
local_size=local_size,
alpha=alpha,
beta=beta,
k=k,
in_place=True)
def caffenet(data,
label=None,
train=True,
num_classes=1000,
classifier_name='fc8',
learn_all=False):
n = caffe.NetSpec()
n.data = data
param = learned_param if learn_all else frozen_param
n.conv1, n.relu1 = conv_relu(n.data, 11, 96, stride=4, param=param)
n.pool1 = max_pool(n.relu1, 3, stride=2)
n.norm1 = L.LRN(n.pool1, local_size=5, alpha=1e-4, beta=0.75)
n.conv2, n.relu2 = conv_relu(n.norm1, 5, 256, pad=2, group=2, param=param)
n.pool2 = max_pool(n.relu2, 3, stride=2)
n.norm2 = L.LRN(n.pool2, local_size=5, alpha=1e-4, beta=0.75)
n.conv3, n.relu3 = conv_relu(n.norm2, 3, 384, pad=1, param=param)
n.conv4, n.relu4 = conv_relu(n.relu3, 3, 384, pad=1, group=2, param=param)
n.conv5, n.relu5 = conv_relu(n.relu4, 3, 256, pad=1, group=2, param=param)
n.pool5 = max_pool(n.relu5, 3, stride=2)
n.fc6, n.relu6 = fc_relu(n.pool5, 4096, param=param)
if train:
n.drop6 = fc7input = L.Dropout(n.relu6, in_place=True)
else:
fc7input = n.relu6
n.fc7, n.relu7 = fc_relu(fc7input, 4096, param=param)
if train:
n.drop7 = fc8input = L.Dropout(n.relu7, in_place=True)
else:
fc8input = n.relu7
# always learn fc8 (param=learned_param)
fc8 = L.InnerProduct(fc8input, num_output=num_classes, param=learned_param)
# give fc8 the name specified by argument `classifier_name`
n.__setattr__(classifier_name, fc8)
if not train:
n.probs = L.Softmax(fc8)
if label is not None:
n.label = label
n.loss = L.SoftmaxWithLoss(fc8, n.label)
n.acc = L.Accuracy(fc8, n.label)
mode = 'train' if train else 'test'
filename = path_savemodel + num_fold + '_' + typeNet + '_' + mode + '.prototxt'
with open(filename, 'w') as f:
f.write(str(n.to_proto()))
return filename
def test_lrn2(self):
n = caffe.NetSpec()
n.input1 = L.Input(shape=make_shape([6, 4, 64, 64]))
n.bnll1 = L.LRN(n.input1,
local_size=7,
alpha=1.1,
beta=0.8,
norm_region=P.LRN.WITHIN_CHANNEL)
self._test_model(*self._netspec_to_model(n, 'lrn2'))
def conv_bn(bottom, nout, ks = 3, stride=1, pad = 0, learn = True):
if learn:
param = [dict(lr_mult=1, decay_mult=1), dict(lr_mult=2, decay_mult=0)]
else:
param = [dict(lr_mult=0, decay_mult=0), dict(lr_mult=0, decay_mult=0)]
conv = L.Convolution(bottom, kernel_size=ks, stride=stride,
num_output=nout, pad=pad, param = param, weight_filler=dict(type="msra"), bias_filler=dict(type="constant"))
bn = L.BatchNorm(conv)
lrn = L.LRN(bn)
return conv, bn, lrn