def __call__(self, x, t):
self.clear()
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv1(x))), 3, stride=2)
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv2(h))), 3, stride=2)
h = F.relu(self.conv3(h))
h = F.relu(self.conv4(h))
h = F.max_pooling_2d(F.relu(self.conv5(h)), 3, stride=2)
h = F.dropout(F.relu(self.fc6(h)), train=self.train)
h = F.dropout(F.relu(self.fc7(h)), train=self.train)
h = self.fc8(h)
self.loss = F.softmax_cross_entropy(h, t)
self.accuracy = F.accuracy(h, t)
return self.loss
python类local_response_normalization()的实例源码
def __call__(self, x, t):
self.clear()
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv1(x))), 3, stride=2)
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv2(h))), 3, stride=2)
h = F.relu(self.conv3(h))
h = F.relu(self.conv4(h))
h = F.max_pooling_2d(F.relu(self.conv5(h)), 3, stride=2)
h = F.dropout(F.relu(self.fc6(h)), train=self.train)
h = F.dropout(F.relu(self.fc7(h)), train=self.train)
h = self.fc8(h)
self.loss = F.softmax_cross_entropy(h, t)
self.accuracy = F.accuracy(h, t)
return self.loss
test_local_response_normalization.py 文件源码
项目：chainer-deconv
作者: germanRos
项目源码
文件源码
阅读 22
收藏 0
点赞 0
评论 0
def check_forward(self, x_data):
x = chainer.Variable(x_data)
y = functions.local_response_normalization(x)
self.assertEqual(y.data.dtype, self.dtype)
y_data = cuda.to_cpu(y.data)
# Naive implementation
y_expect = numpy.zeros_like(self.x)
for n, c, h, w in numpy.ndindex(self.x.shape):
s = 0
for i in six.moves.range(max(0, c - 2), min(7, c + 2)):
s += self.x[n, i, h, w] ** 2
denom = (2 + 1e-4 * s) ** .75
y_expect[n, c, h, w] = self.x[n, c, h, w] / denom
gradient_check.assert_allclose(
y_expect, y_data, **self.check_forward_optionss)
def __call__(self, x, depth=1):
assert 1 <= depth <= self.n_encdec
h = F.local_response_normalization(x, 5, 1, 0.0005, 0.75)
# Unchain the inner EncDecs after the given depth
encdec = getattr(self, 'encdec{}'.format(depth))
encdec.inside = None
h = self.encdec1(h, train=self.train)
h = self.conv_cls(h)
return h
def __call__(self, x, y, t):
self.clear()
hR = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.convR1(x))), 3, stride=2)
hR = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.convR2(hR))), 3, stride=2)
hR = F.relu(self.convR3(hR))
hR = F.relu(self.convR4(hR))
hR = F.max_pooling_2d(F.relu(self.convR5(hR)), 3, stride=2)
hR = F.dropout(F.relu(self.fcR6(hR)), train=self.train)
hR = F.dropout(F.relu(self.fcR7(hR)), train=self.train)
hD = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.convD1(y))), 3, stride=2)
hD = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.convD2(hD))), 3, stride=2)
hD = F.relu(self.convD3(hD))
hD = F.relu(self.convD4(hD))
hD = F.max_pooling_2d(F.relu(self.convD5(hD)), 3, stride=2)
hD = F.dropout(F.relu(self.fcD6(hD)), train=self.train)
hD = F.dropout(F.relu(self.fcD7(hD)), train=self.train)
h = F.dropout(F.relu(self.fc8(hR, hD)), train=self.train)
h = self.fc9(h)
self.loss = F.softmax_cross_entropy(h, t)
self.accuracy = F.accuracy(h, t)
return self.loss
def _setup_lrn(self, layer):
param = layer.lrn_param
if param.norm_region != param.ACROSS_CHANNELS:
raise RuntimeError('Within-channel LRN is not supported')
fwd = _SingleArgumentFunction(
functions.local_response_normalization,
n=param.local_size, k=param.k,
alpha=param.alpha / param.local_size, beta=param.beta)
self.forwards[layer.name] = fwd
self._add_layer(layer)
def __call__(self, x, subtract_mean=True):
if subtract_mean:
x = x - self._image_mean
# h = super(ModifiedGoogLeNet, self).__call__(
# x, layers=['pool5'], train=train)['pool5']
# h = self.bn_fc(h, test=not train)
# y = self.fc(h)
# return y
h = F.relu(self.conv1(x))
h = F.max_pooling_2d(h, 3, stride=2)
h = F.local_response_normalization(h, n=5, k=1, alpha=1e-4/5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.local_response_normalization(h, n=5, k=1, alpha=1e-4/5)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.bn_fc(h)
y = self.fc(h)
if self.normalize_output:
y = F.normalize(y)
return y
def __call__(self, x):
"""Compute an image-wise score from a batch of images
Args:
x (chainer.Variable): A variable with 4D image array.
Returns:
chainer.Variable:
An image-wise score. Its channel size is :obj:`self.n_class`.
"""
p1 = F.MaxPooling2D(2, 2)
p2 = F.MaxPooling2D(2, 2)
p3 = F.MaxPooling2D(2, 2)
p4 = F.MaxPooling2D(2, 2)
h = F.local_response_normalization(x, 5, 1, 1e-4 / 5., 0.75)
h = _pool_without_cudnn(p1, F.relu(self.conv1_bn(self.conv1(h))))
h = _pool_without_cudnn(p2, F.relu(self.conv2_bn(self.conv2(h))))
h = _pool_without_cudnn(p3, F.relu(self.conv3_bn(self.conv3(h))))
h = _pool_without_cudnn(p4, F.relu(self.conv4_bn(self.conv4(h))))
h = self._upsampling_2d(h, p4)
h = self.conv_decode4_bn(self.conv_decode4(h))
h = self._upsampling_2d(h, p3)
h = self.conv_decode3_bn(self.conv_decode3(h))
h = self._upsampling_2d(h, p2)
h = self.conv_decode2_bn(self.conv_decode2(h))
h = self._upsampling_2d(h, p1)
h = self.conv_decode1_bn(self.conv_decode1(h))
score = self.conv_classifier(h)
return score
def _setup_lrn(self, layer):
param = layer.lrn_param
if param.norm_region != param.ACROSS_CHANNELS:
raise RuntimeError('Within-channel LRN is not supported')
fwd = _SingleArgumentFunction(
functions.local_response_normalization,
n=param.local_size, k=param.k,
alpha=param.alpha / param.local_size, beta=param.beta)
self.forwards[layer.name] = fwd
self._add_layer(layer)
def forward(self, x):
h = F.relu(self.conv1(x))
h = F.local_response_normalization(
F.max_pooling_2d(h, 3, stride=2), n=5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.max_pooling_2d(
F.local_response_normalization(h, n=5), 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss1_conv(l))
l = F.relu(self.loss1_fc1(l))
l = self.loss1_fc2(l)
loss1 = l
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss2_conv(l))
l = F.relu(self.loss2_fc1(l))
l = self.loss2_fc2(l)
loss2 = l
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.loss3_fc(F.dropout(h, 0.4, train=self.train))
loss3 = h
return loss1,loss2,loss3
def forward(self, x):
y1 = self.model['conv1/7x7_s2'](x)
h = F.relu(y1)
h = F.local_response_normalization(self.pool_func(h, 3, stride=2), n=5)
h = F.relu(self.model['conv2/3x3_reduce'](h))
y2 = self.model['conv2/3x3'](h)
h = F.relu(y2)
h = self.pool_func(F.local_response_normalization(h, n=5), 3, stride=2)
out1 = self.model['inception_3a/1x1'](h)
out3 = self.model['inception_3a/3x3'](F.relu(self.model['inception_3a/3x3_reduce'](h)))
out5 = self.model['inception_3a/5x5'](F.relu(self.model['inception_3a/5x5_reduce'](h)))
pool = self.model['inception_3a/pool_proj'](self.pool_func(h, 3, stride=1, pad=1))
y3 = F.concat((out1, out3, out5, pool), axis=1)
h = F.relu(y3)
out1 = self.model['inception_3b/1x1'](h)
out3 = self.model['inception_3b/3x3'](F.relu(self.model['inception_3b/3x3_reduce'](h)))
out5 = self.model['inception_3b/5x5'](F.relu(self.model['inception_3b/5x5_reduce'](h)))
pool = self.model['inception_3b/pool_proj'](self.pool_func(h, 3, stride=1, pad=1))
y4 = F.concat((out1, out3, out5, pool), axis=1)
h = F.relu(y4)
h = self.pool_func(h, 3, stride=2)
out1 = self.model['inception_4a/1x1'](h)
out3 = self.model['inception_4a/3x3'](F.relu(self.model['inception_4a/3x3_reduce'](h)))
out5 = self.model['inception_4a/5x5'](F.relu(self.model['inception_4a/5x5_reduce'](h)))
pool = self.model['inception_4a/pool_proj'](self.pool_func(h, 3, stride=1, pad=1))
y5 = F.concat((out1, out3, out5, pool), axis=1)
h = F.relu(y5)
out1 = self.model['inception_4b/1x1'](h)
out3 = self.model['inception_4b/3x3'](F.relu(self.model['inception_4b/3x3_reduce'](h)))
out5 = self.model['inception_4b/5x5'](F.relu(self.model['inception_4b/5x5_reduce'](h)))
pool = self.model['inception_4b/pool_proj'](self.pool_func(h, 3, stride=1, pad=1))
y6 = F.concat((out1, out3, out5, pool), axis=1)
h = F.relu(y6)
return [y1,y2,y3,y4,y5,y6]
def predict(self, x):
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv1(x))), 3, stride=2)
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv2(h))), 3, stride=2)
h = F.relu(self.conv3(h))
h = F.relu(self.conv4(h))
h = F.max_pooling_2d(F.relu(self.conv5(h)), 3, stride=2)
h = F.dropout(F.relu(self.fc6(h)), train=self.train)
h = F.dropout(F.relu(self.fc7(h)), train=self.train)
h = F.dropout(F.relu(self.fc8(h)), train=self.train)
h = self.fc9(h)
return h
def l2norm(self, x):
"""
Force embeddings onto a hypershere.
This has been done in
Schroff et al. “FaceNet: A Unified Embedding for Face Recognition and
Clustering.” arXiv:1503.03832 [cs], March 12, 2015.
http://arxiv.org/abs/1503.03832.
However it did not work out for me very well, so this function is
currently not used.
"""
return F.local_response_normalization(x, n=x.data.shape[1]*2,
k=0, alpha=1, beta=0.5)
def __call__(self, x, t):
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv1(x))), 3, stride=2)
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv2(h))), 3, stride=2)
h = F.relu(self.conv3(h))
h = F.relu(self.conv4(h))
h = F.max_pooling_2d(F.relu(self.conv5(h)), 3, stride=1)
h = F.dropout(F.relu(self.fc6(h)), train=self.train)
h = F.dropout(F.relu(self.fc7(h)), train=self.train)
h = self.fc8(h)
self.loss = F.softmax_cross_entropy(h, t)
self.accuracy = F.accuracy(h, t)
return self.loss
def __call__(self, x, t):
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv1(x))), 2, stride=2)
h = F.max_pooling_2d(F.relu(
F.local_response_normalization(self.conv2(h))), 2, stride=2)
h = F.dropout(F.relu(self.conv3(h)), ratio=0.7, train=self.train)
h = F.max_pooling_2d(F.relu(self.conv4(h)), 2, stride=2)
h = F.max_pooling_2d(F.relu(self.conv5(h)), 2, stride=2, cover_all=True)
h = F.dropout(F.relu(self.fc6(h)), ratio=0.7, train=self.train)
h = F.dropout(F.relu(self.fc7(h)), ratio=0.7, train=self.train)
h = self.fc8(h)
self.loss = F.softmax_cross_entropy(h, t)
self.accuracy = F.accuracy(h, t)
return self.loss
def __call__(self, x, t):
self.clear()
h = F.relu(self.conv1(x))
h = F.local_response_normalization(
F.max_pooling_2d(h, 3, stride=2), n=5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.max_pooling_2d(
F.local_response_normalization(h, n=5), 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss1_conv(l))
l = F.relu(self.loss1_fc1(l))
l = self.loss1_fc2(l)
self.loss1 = F.softmax_cross_entropy(l, t)
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss2_conv(l))
l = F.relu(self.loss2_fc1(l))
l = self.loss2_fc2(l)
self.loss2 = F.softmax_cross_entropy(l, t)
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.loss3_fc(F.dropout(h, 0.4, train=self.train))
self.loss3 = F.softmax_cross_entropy(h, t)
self.loss = 0.3 * (self.loss1 + self.loss2) + self.loss3
self.accuracy = F.accuracy(h, t)
return self.loss
def __call__(self, x, t):
h = F.relu(self.conv1(x))
h = F.local_response_normalization(
F.max_pooling_2d(h, 3, stride=2), n=5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.max_pooling_2d(
F.local_response_normalization(h, n=5), 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss1_conv(l))
l = F.relu(self.loss1_fc1(l))
l = self.loss1_fc2(l)
loss1 = F.softmax_cross_entropy(l, t)
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss2_conv(l))
l = F.relu(self.loss2_fc1(l))
l = self.loss2_fc2(l)
loss2 = F.softmax_cross_entropy(l, t)
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.loss3_fc(F.dropout(h, 0.4, train=self.train))
loss3 = F.softmax_cross_entropy(h, t)
loss = 0.3 * (loss1 + loss2) + loss3
accuracy = F.accuracy(h, t)
chainer.report({
'loss': loss,
'loss1': loss1,
'loss2': loss2,
'loss3': loss3,
'accuracy': accuracy
}, self)
return loss
def __call__(self, x, t):
h = F.relu(self.conv1(x))
h = F.local_response_normalization(
F.max_pooling_2d(h, 3, stride=2), n=5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.max_pooling_2d(
F.local_response_normalization(h, n=5), 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss1_conv(l))
l = F.relu(self.loss1_fc1(l))
l = self.loss1_fc2(l)
loss1 = F.softmax_cross_entropy(l, t)
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss2_conv(l))
l = F.relu(self.loss2_fc1(l))
l = self.loss2_fc2(l)
loss2 = F.softmax_cross_entropy(l, t)
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.loss3_fc(F.dropout(h, 0.4))
loss3 = F.softmax_cross_entropy(h, t)
loss = 0.3 * (loss1 + loss2) + loss3
accuracy = F.accuracy(h, t)
chainer.report({
'loss': loss,
'loss1': loss1,
'loss2': loss2,
'loss3': loss3,
'accuracy': accuracy
}, self)
return loss
def __call__(self, x, t):
h = F.relu(self.conv1(x))
h = F.local_response_normalization(
F.max_pooling_2d(h, 3, stride=2), n=5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.max_pooling_2d(
F.local_response_normalization(h, n=5), 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss1_conv(l))
l = F.relu(self.loss1_fc1(l))
l = self.loss1_fc2(l)
loss1 = F.softmax_cross_entropy(l, t)
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss2_conv(l))
l = F.relu(self.loss2_fc1(l))
l = self.loss2_fc2(l)
loss2 = F.softmax_cross_entropy(l, t)
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.loss3_fc(F.dropout(h, 0.4, train=self.train))
loss3 = F.softmax_cross_entropy(h, t)
loss = 0.3 * (loss1 + loss2) + loss3
accuracy = F.accuracy(h, t)
chainer.report({
'loss': loss,
'loss1': loss1,
'loss2': loss2,
'loss3': loss3,
'accuracy': accuracy
}, self)
return loss
def __call__(self, x, t):
h = F.relu(self.conv1(x))
h = F.local_response_normalization(
F.max_pooling_2d(h, 3, stride=2), n=5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.max_pooling_2d(
F.local_response_normalization(h, n=5), 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss1_conv(l))
l = F.relu(self.loss1_fc1(l))
l = self.loss1_fc2(l)
loss1 = F.softmax_cross_entropy(l, t)
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss2_conv(l))
l = F.relu(self.loss2_fc1(l))
l = self.loss2_fc2(l)
loss2 = F.softmax_cross_entropy(l, t)
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.loss3_fc(F.dropout(h, 0.4))
loss3 = F.softmax_cross_entropy(h, t)
loss = 0.3 * (loss1 + loss2) + loss3
return loss
def __call__(self, x, t):
h = F.relu(self.conv1(x))
h = F.local_response_normalization(
F.max_pooling_2d(h, 3, stride=2), n=5)
h = F.relu(self.conv2_reduce(h))
h = F.relu(self.conv2(h))
h = F.max_pooling_2d(
F.local_response_normalization(h, n=5), 3, stride=2)
h = self.inc3a(h)
h = self.inc3b(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc4a(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss1_conv(l))
l = F.relu(self.loss1_fc1(l))
l = self.loss1_fc2(l)
loss1 = F.softmax_cross_entropy(l, t)
h = self.inc4b(h)
h = self.inc4c(h)
h = self.inc4d(h)
l = F.average_pooling_2d(h, 5, stride=3)
l = F.relu(self.loss2_conv(l))
l = F.relu(self.loss2_fc1(l))
l = self.loss2_fc2(l)
loss2 = F.softmax_cross_entropy(l, t)
h = self.inc4e(h)
h = F.max_pooling_2d(h, 3, stride=2)
h = self.inc5a(h)
h = self.inc5b(h)
h = F.average_pooling_2d(h, 7, stride=1)
h = self.loss3_fc(F.dropout(h, 0.4))
loss3 = F.softmax_cross_entropy(h, t)
loss = 0.3 * (loss1 + loss2) + loss3
accuracy = F.accuracy(h, t)
chainer.report({
'loss': loss,
'loss1': loss1,
'loss2': loss2,
'loss3': loss3,
'accuracy': accuracy
}, self)
return loss
