def __init__(self, layers, init_img, subject_img, style_img,
subject_weights, style_weights, smoothness=0.0):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l+1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers]
# Setup network
x_shape = init_img.shape
self.x = Parameter(init_img)
self.x.setup(x_shape)
for layer in self.layers:
layer.setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None]*len(self.layers)
self.style_grams = [None]*len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
self.tv_weight = smoothness
kernel = np.array([[0, 1, 0], [1, -4, 1], [0, 1, 0]], dtype=dp.float_)
kernel /= np.sum(np.abs(kernel))
self.tv_kernel = ca.array(kernel[np.newaxis, np.newaxis, ...])
self.tv_conv = ca.nnet.ConvBC01((1, 1), (1, 1))
def __init__(self, n_hidden, n_out, w_xh=0, w_hh=0, w_hy=0, bias_h=0.0, bias_y=0, activation="tanh"):
self.name = "recurrent"
self.n_hidden = n_hidden
self.n_out = n_out
self.w_xh = Parameter.from_any(w_xh)
self.w_hh = Parameter.from_any(w_hh)
self.w_hy = Parameter.from_any(w_hy)
self.b_h = Parameter.from_any(bias_h)
self.b_y = Parameter.from_any(bias_y)
self.activation = Activation.from_any(activation)
self._tmp_x = None
def __init__(self, n_hidden, n_out, w_xh=0, w_hh=0, w_hy=0, bias_h=0.0,
bias_y=0, activation='tanh'):
self.name = 'recurrent'
self.n_hidden = n_hidden
self.n_out = n_out
self.w_xh = Parameter.from_any(w_xh)
self.w_hh = Parameter.from_any(w_hh)
self.w_hy = Parameter.from_any(w_hy)
self.b_h = Parameter.from_any(bias_h)
self.b_y = Parameter.from_any(bias_y)
self.activation = Activation.from_any(activation)
self._tmp_x = None
def __init__(self, layers, subject_img, style_img, subject_weights,
style_weights):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l + 1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [
Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers
]
# Setup network
x_shape = subject_img.shape
self.x = Parameter(subject_img)
self.x._setup(x_shape)
for layer in self.layers:
layer._setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None] * len(self.layers)
self.style_grams = [None] * len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
class GramNet(Model):
def __init__(self, layers, style_weights):
# Map weights (in convolution indices) to layer indices
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.style_weights[l] = style_weights[conv_idx]
if style_weights[conv_idx] > 0:
layers_len = l+1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [Convolution(l) if isinstance(l, dp.Convolution) else l for l in layers]
def compute_grams(self, image):
# Setup network
x_shape = image.shape
self.x = Parameter(image)
self.x._setup(x_shape)
for layer in self.layers:
layer._setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.style_grams = [None]*len(self.layers)
next_style = ca.array(image)
for l, layer in enumerate(self.layers):
next_style = layer.fprop(next_style)
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
self.style_grams[l] = gram
return self.style_grams;
@property
def image(self):
return np.array(self.x.array)
@property
def _params(self):
return [self.x]
def _update(self):
raise NotImplementedError("This network can only compute the stlye gram matrix.")
def __init__(self, n_hidden, w_x, w_h, bias_filler=0):
self.name = "gru"
self.n_hidden = n_hidden
self.w_x = Parameter.from_any(w_x)
self.w_h = Parameter.from_any(w_h)
self.b_r = Parameter.from_any(bias_filler)
self.b_u = Parameter.from_any(bias_filler)
self.b_c = Parameter.from_any(bias_filler)
self.act_r = Activation.from_any("sigmoid")
self.act_u = Activation.from_any("sigmoid")
self.act_c = Activation.from_any("tanh")
self.clip = 5
self._tmp_x = None
def __init__(self, n_hidden, w_x, w_h, bias_filler=0):
self.name = 'gru'
self.n_hidden = n_hidden
self.w_x = Parameter.from_any(w_x)
self.w_h = Parameter.from_any(w_h)
self.b_r = Parameter.from_any(bias_filler)
self.b_u = Parameter.from_any(bias_filler)
self.b_c = Parameter.from_any(bias_filler)
self.act_r = Activation.from_any('sigmoid')
self.act_u = Activation.from_any('sigmoid')
self.act_c = Activation.from_any('tanh')
self.clip = 5
self._tmp_x = None
def __init__(self, layers, init_img, subject_img, style_img,
subject_weights, style_weights, smoothness=0.0):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l+1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers]
# Setup network
x_shape = init_img.shape
self.x = Parameter(init_img)
self.x.setup(x_shape)
for layer in self.layers:
layer.setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None]*len(self.layers)
self.style_grams = [None]*len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
self.tv_weight = smoothness
kernel = np.array([[0, 1, 0], [1, -4, 1], [0, 1, 0]], dtype=dp.float_)
kernel /= np.sum(np.abs(kernel))
self.tv_kernel = ca.array(kernel[np.newaxis, np.newaxis, ...])
self.tv_conv = ca.nnet.ConvBC01((1, 1), (1, 1))
def __init__(self, layers, subject_img, style_img, subject_weights,
style_weights):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l+1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers]
# Setup network
x_shape = subject_img.shape
self.x = Parameter(subject_img)
self.x._setup(x_shape)
for layer in self.layers:
layer._setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None]*len(self.layers)
self.style_grams = [None]*len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
def compute_grams(self, image):
# Setup network
x_shape = image.shape
self.x = Parameter(image)
self.x._setup(x_shape)
for layer in self.layers:
layer._setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.style_grams = [None]*len(self.layers)
next_style = ca.array(image)
for l, layer in enumerate(self.layers):
next_style = layer.fprop(next_style)
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
self.style_grams[l] = gram
return self.style_grams;
class StyleNetwork(Model):
""" Artistic style network
Implementation of [1].
Differences:
- The gradients for both subject and style are normalized. The original
method uses pre-normalized convolutional features.
- The Gram matrices are scaled wrt. # of pixels. The original method is
sensitive to different image sizes between subject and style.
- Additional smoothing term for visually better results.
References:
[1]: A Neural Algorithm of Artistic Style; Leon A. Gatys, Alexander S.
Ecker, Matthias Bethge; arXiv:1508.06576; 08/2015
"""
def __init__(self,
layers,
init_img,
subject_img,
style_img,
subject_weights,
style_weights,
smoothness=0.0):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l + 1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [
Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers
]
# Setup network
x_shape = init_img.shape
self.x = Parameter(init_img)
self.x._setup(x_shape)
for layer in self.layers:
layer._setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None] * len(self.layers)
self.style_grams = [None] * len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
self.tv_weight = smoothness
kernel = np.array([[0, 1, 0], [1, -4, 1], [0, 1, 0]], dtype=dp.float_)
kernel /= np.sum(np.abs(kernel))
self.tv_kernel = ca.array(kernel[np.newaxis, np.newaxis, ...])
self.tv_conv = ca.nnet.ConvBC01((1, 1), (1, 1))
@property
def image(self):
return np.array(self.x.array)
@property
def _params(self):
return [self.x]
def _update(self):
# Forward propagation
next_x = self.x.array
x_feats = [None] * len(self.layers)
for l, layer in enumerate(self.layers):
next_x = layer.fprop(next_x)
if self.subject_weights[l] > 0 or self.style_weights[l] > 0:
x_feats[l] = next_x
# Backward propagation
grad = ca.zeros_like(next_x)
loss = ca.zeros(1)
for l, layer in reversed(list(enumerate(self.layers))):
if self.subject_weights[l] > 0:
diff = x_feats[l] - self.subject_feats[l]
norm = ca.sum(ca.fabs(diff)) + 1e-8
weight = float(self.subject_weights[l]) / norm
grad += diff * weight
loss += 0.5 * weight * ca.sum(diff**2)
if self.style_weights[l] > 0:
diff = gram_matrix(x_feats[l]) - self.style_grams[l]
n_channels = diff.shape[0]
x_feat = ca.reshape(x_feats[l], (n_channels, -1))
style_grad = ca.reshape(ca.dot(diff, x_feat), x_feats[l].shape)
norm = ca.sum(ca.fabs(style_grad))
weight = float(self.style_weights[l]) / norm
style_grad *= weight
grad += style_grad
loss += 0.25 * weight * ca.sum(diff**2)
grad = layer.bprop(grad)
if self.tv_weight > 0:
x = ca.reshape(self.x.array, (3, 1) + grad.shape[2:])
tv = self.tv_conv.fprop(x, self.tv_kernel)
tv *= self.tv_weight
grad -= ca.reshape(tv, grad.shape)
ca.copyto(self.x.grad_array, grad)
return loss
class StyleNetwork(Model):
""" Artistic style network
Implementation of [1].
Differences:
- The gradients for both subject and style are normalized. The original
gradient summatino scheme seemed sensitive to changes in image size.
References:
[1]: A Neural Algorithm of Artistic Style; Leon A. Gatys, Alexander S.
Ecker, Matthias Bethge; arXiv:1508.06576; 08/2015
"""
def __init__(self, layers, subject_img, style_img, subject_weights,
style_weights):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l + 1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [
Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers
]
# Setup network
x_shape = subject_img.shape
self.x = Parameter(subject_img)
self.x._setup(x_shape)
for layer in self.layers:
layer._setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None] * len(self.layers)
self.style_grams = [None] * len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
@property
def image(self):
return np.array(self.x.array)
@property
def _params(self):
return [self.x]
def _update(self):
# Forward propagation
next_x = self.x.array
x_feats = [None] * len(self.layers)
x_grams = [None] * len(self.layers)
for l, layer in enumerate(self.layers):
next_x = layer.fprop(next_x)
if self.subject_weights[l] > 0:
x_feats[l] = next_x
if self.style_weights[l] > 0:
x_feats[l] = next_x
x_grams[l] = gram_matrix(next_x)
# Backward propagation
grad = ca.zeros_like(next_x)
loss = ca.zeros(1)
for l, layer in reversed(list(enumerate(self.layers))):
if self.subject_weights[l] > 0:
diff = x_feats[l] - self.subject_feats[l]
norm = ca.sum(ca.fabs(diff)) + 1e-8
weight = float(self.subject_weights[l]) / norm
grad += diff * weight
loss += 0.5 * weight * ca.sum(diff**2)
if self.style_weights[l] > 0:
diff = x_grams[l] - self.style_grams[l]
n_channels = diff.shape[0]
x_feat = ca.reshape(x_feats[l], (n_channels, -1))
style_grad = ca.reshape(ca.dot(diff, x_feat), x_feats[l].shape)
norm = ca.sum(ca.fabs(style_grad))
weight = float(self.style_weights[l]) / norm
style_grad *= weight
grad += style_grad
loss += 0.25 * weight * ca.sum(diff**2)
grad = layer.bprop(grad)
ca.copyto(self.x.grad_array, grad)
return loss
class StyleNetwork(Model):
""" Artistic style network
Implementation of [1].
Differences:
- The gradients for both subject and style are normalized. The original
method uses pre-normalized convolutional features.
- The Gram matrices are scaled wrt. # of pixels. The original method is
sensitive to different image sizes between subject and style.
- Additional smoothing term for visually better results.
References:
[1]: A Neural Algorithm of Artistic Style; Leon A. Gatys, Alexander S.
Ecker, Matthias Bethge; arXiv:1508.06576; 08/2015
"""
def __init__(self, layers, init_img, subject_img, style_img,
subject_weights, style_weights, smoothness=0.0):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l+1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers]
# Setup network
x_shape = init_img.shape
self.x = Parameter(init_img)
self.x.setup(x_shape)
for layer in self.layers:
layer.setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None]*len(self.layers)
self.style_grams = [None]*len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
self.tv_weight = smoothness
kernel = np.array([[0, 1, 0], [1, -4, 1], [0, 1, 0]], dtype=dp.float_)
kernel /= np.sum(np.abs(kernel))
self.tv_kernel = ca.array(kernel[np.newaxis, np.newaxis, ...])
self.tv_conv = ca.nnet.ConvBC01((1, 1), (1, 1))
@property
def image(self):
return np.array(self.x.array)
@property
def params(self):
return [self.x]
def update(self):
# Forward propagation
next_x = self.x.array
x_feats = [None]*len(self.layers)
for l, layer in enumerate(self.layers):
next_x = layer.fprop(next_x)
if self.subject_weights[l] > 0 or self.style_weights[l] > 0:
x_feats[l] = next_x
# Backward propagation
grad = ca.zeros_like(next_x)
loss = ca.zeros(1)
for l, layer in reversed(list(enumerate(self.layers))):
if self.subject_weights[l] > 0:
diff = x_feats[l] - self.subject_feats[l]
norm = ca.sum(ca.fabs(diff)) + 1e-8
weight = float(self.subject_weights[l]) / norm
grad += diff * weight
loss += 0.5*weight*ca.sum(diff**2)
if self.style_weights[l] > 0:
diff = gram_matrix(x_feats[l]) - self.style_grams[l]
n_channels = diff.shape[0]
x_feat = ca.reshape(x_feats[l], (n_channels, -1))
style_grad = ca.reshape(ca.dot(diff, x_feat), x_feats[l].shape)
norm = ca.sum(ca.fabs(style_grad))
weight = float(self.style_weights[l]) / norm
style_grad *= weight
grad += style_grad
loss += 0.25*weight*ca.sum(diff**2)
grad = layer.bprop(grad)
if self.tv_weight > 0:
x = ca.reshape(self.x.array, (3, 1) + grad.shape[2:])
tv = self.tv_conv.fprop(x, self.tv_kernel)
tv *= self.tv_weight
grad -= ca.reshape(tv, grad.shape)
ca.copyto(self.x.grad_array, grad)
return loss
class StyleNetwork(Model):
""" Artistic style network
Implementation of [1].
Differences:
- The gradients for both subject and style are normalized. The original
gradient summatino scheme seemed sensitive to changes in image size.
References:
[1]: A Neural Algorithm of Artistic Style; Leon A. Gatys, Alexander S.
Ecker, Matthias Bethge; arXiv:1508.06576; 08/2015
"""
def __init__(self, layers, subject_img, style_img, subject_weights,
style_weights):
# Map weights (in convolution indices) to layer indices
self.subject_weights = np.zeros(len(layers))
self.style_weights = np.zeros(len(layers))
layers_len = 0
conv_idx = 0
for l, layer in enumerate(layers):
if isinstance(layer, dp.Activation):
self.subject_weights[l] = subject_weights[conv_idx]
self.style_weights[l] = style_weights[conv_idx]
if subject_weights[conv_idx] > 0 or \
style_weights[conv_idx] > 0:
layers_len = l+1
conv_idx += 1
# Discard unused layers
layers = layers[:layers_len]
# Wrap convolution layers for better performance
self.layers = [Convolution(l) if isinstance(l, dp.Convolution) else l
for l in layers]
# Setup network
x_shape = subject_img.shape
self.x = Parameter(subject_img)
self.x._setup(x_shape)
for layer in self.layers:
layer._setup(x_shape)
x_shape = layer.y_shape(x_shape)
# Precompute subject features and style Gram matrices
self.subject_feats = [None]*len(self.layers)
self.style_grams = [None]*len(self.layers)
next_subject = ca.array(subject_img)
next_style = ca.array(style_img)
for l, layer in enumerate(self.layers):
next_subject = layer.fprop(next_subject)
next_style = layer.fprop(next_style)
if self.subject_weights[l] > 0:
self.subject_feats[l] = next_subject
if self.style_weights[l] > 0:
gram = gram_matrix(next_style)
# Scale gram matrix to compensate for different image sizes
n_pixels_subject = np.prod(next_subject.shape[2:])
n_pixels_style = np.prod(next_style.shape[2:])
scale = (n_pixels_subject / float(n_pixels_style))
self.style_grams[l] = gram * scale
@property
def image(self):
return np.array(self.x.array)
@property
def _params(self):
return [self.x]
def _update(self):
# Forward propagation
next_x = self.x.array
x_feats = [None]*len(self.layers)
x_grams = [None]*len(self.layers)
for l, layer in enumerate(self.layers):
next_x = layer.fprop(next_x)
if self.subject_weights[l] > 0:
x_feats[l] = next_x
if self.style_weights[l] > 0:
x_feats[l] = next_x
x_grams[l] = gram_matrix(next_x)
# Backward propagation
grad = ca.zeros_like(next_x)
loss = ca.zeros(1)
for l, layer in reversed(list(enumerate(self.layers))):
if self.subject_weights[l] > 0:
diff = x_feats[l] - self.subject_feats[l]
norm = ca.sum(ca.fabs(diff)) + 1e-8
weight = float(self.subject_weights[l]) / norm
grad += diff * weight
loss += 0.5*weight*ca.sum(diff**2)
if self.style_weights[l] > 0:
diff = x_grams[l] - self.style_grams[l]
n_channels = diff.shape[0]
x_feat = ca.reshape(x_feats[l], (n_channels, -1))
style_grad = ca.reshape(ca.dot(diff, x_feat), x_feats[l].shape)
norm = ca.sum(ca.fabs(style_grad))
weight = float(self.style_weights[l]) / norm
style_grad *= weight
grad += style_grad
loss += 0.25*weight*ca.sum(diff**2)
grad = layer.bprop(grad)
ca.copyto(self.x.grad_array, grad)
return loss
