pystiche_papers.gatys_ecker_bethge_2016

Title	Image Style Transfer Using Convolutional Neural Networks
Authors	Leon A. Gatys, Alexander. S. Ecker, and Matthias Bethge
Citation	[GEB2016]
Reference implementation	Repository / Archive
Variant	Image optimization
Content loss	FeatureReconstructionLoss
Style loss	GramLoss

Behavioral changes

See also

Paper implementations

The following parts are affected:

• FeatureReconstructionLoss

• MultiLayerEncodingLoss

• multi_layer_encoder

Hyper parameters

See also

Paper implementations

Empty cells mean, that the parameter is not defined in the paper or no default is set in the reference implementation of the original authors. In both cases the available value is used as default.

content_loss()

Parameter	impl_params
	True	False
layer	"relu4_2"	"conv4_2"
score_weight	1e0

style_loss()

Parameter	impl_params
	True		False
layers | ("relu1_1", "relu2_1", "relu3_1", "relu4_1", "relu5_1")			("conv1_1", "conv2_1", "conv3_1", "conv4_1", "conv5_1")
layer_weights		(2.4e-04, 6.1e-05, 1.5e-05, 3.8e-06, 3.8e-06) 1	"mean"
score_weight		1e3 2

nst()

Parameter	impl_params
	True	False
image_size	512
starting_point	"content"	"random"
num_steps	500

1

The layer_weights are computed by \(1 / n^2\) where \(n\) denotes the number of channels of a feature map from the corresponding layer in the multi_layer_encoder().

2

The paper also reports score_weight=1e-4 for some images.

API

pystiche_papers.gatys_ecker_bethge_2016.images()

Return type

DownloadableImageCollection

class pystiche_papers.gatys_ecker_bethge_2016.FeatureReconstructionLoss(encoder, impl_params=True, **feature_reconstruction_loss_kwargs)

Feature reconstruction loss from [GEB2016].

Parameters

• encoder (Encoder) – Encoder used to encode the input.

• impl_params (bool) – If False, calculate the score with the squared error (SE) instead of the mean squared error (MSE). Furthermore, use a score correction factor of 1/2.

• **feature_reconstruction_loss_kwargs – Additional parameters of a pystiche.loss.FeatureReconstructionLoss.

See also

• pystiche.loss.FeatureReconstructionLoss

pystiche_papers.gatys_ecker_bethge_2016.content_loss(impl_params=True, multi_layer_encoder=None, hyper_parameters=None)

Content loss from [GEB2016].

Parameters

• impl_params (bool) – Switch the behavior and hyper-parameters between the reference implementation of the original authors and what is described in the paper. For details see here.

• multi_layer_encoder (Optional[MultiLayerEncoder]) – Pretrained multi-layer encoder. If omitted, multi_layer_encoder() is used.

• hyper_parameters (Optional[HyperParameters]) – Hyper parameters. If omitted, hyper_parameters() is used.

See also

• pystiche_papers.gatys_ecker_bethge_2016.FeatureReconstructionLoss

Return type

FeatureReconstructionLoss

class pystiche_papers.gatys_ecker_bethge_2016.MultiLayerEncodingLoss(multi_layer_encoder, layers, encoding_loss_fn, impl_params=True, **multi_layer_encoding_op_kwargs)

Multi-layer encoding loss from [GEB2016].

Parameters

• multi_layer_encoder (MultiLayerEncoder) – Multi-layer encoder.

• layers (Sequence[str]) – Layers of the multi_layer_encoder that the children losses operate on.

• encoding_loss_fn (Callable[[Encoder, float], Loss]) – Callable that returns a children loss given a pystiche.enc.SingleLayerEncoder extracted from the multi_layer_encoder and its corresponding layer weight.

• impl_params (bool) – If False, use a score correction factor of 1/4.

• **multi_layer_encoding_op_kwargs – Additional parameters of a pystiche.loss.MultiLayerEncodingLoss.

See also

• pystiche.loss.MultiLayerEncodingLoss

pystiche_papers.gatys_ecker_bethge_2016.style_loss(impl_params=True, multi_layer_encoder=None, hyper_parameters=None)

Style loss from [GEB2016].

Parameters

• impl_params (bool) – Switch the behavior and hyper-parameters between the reference implementation of the original authors and what is described in the paper. For details see here.

• multi_layer_encoder (Optional[MultiLayerEncoder]) – Pretrained multi-layer encoder. If omitted, multi_layer_encoder() is used.

• hyper_parameters (Optional[HyperParameters]) – Hyper parameters. If omitted, hyper_parameters() is used.

See also

• pystiche_papers.gatys_ecker_bethge_2016.MultiLayerEncodingLoss

Return type

MultiLayerEncodingLoss

pystiche_papers.gatys_ecker_bethge_2016.perceptual_loss(impl_params=True, multi_layer_encoder=None, hyper_parameters=None)

Perceptual loss from [GEB2016].

Parameters

• impl_params (bool) – Switch the behavior and hyper-parameters between the reference implementation of the original authors and what is described in the paper. For details see here.

• multi_layer_encoder (Optional[MultiLayerEncoder]) – Pretrained multi-layer encoder. If omitted, multi_layer_encoder() is used.

• hyper_parameters (Optional[HyperParameters]) – Hyper parameters. If omitted, hyper_parameters() is used.

See also

• pystiche_papers.gatys_ecker_bethge_2016.content_loss()

• pystiche_papers.gatys_ecker_bethge_2016.style_loss()

Return type

PerceptualLoss

pystiche_papers.gatys_ecker_bethge_2016.nst(content_image, style_image, impl_params=True, hyper_parameters=None, quiet=False)

NST from [GEB2016].

Parameters

• content_image (Tensor) – Content image for the NST.

• style_image (Tensor) – Style image for the NST.

• impl_params (bool) – Switch the behavior and hyper-parameters between the reference implementation of the original authors and what is described in the paper. For details see here.

• hyper_parameters (Optional[HyperParameters]) – If omitted, hyper_parameters() is used.

• quiet (bool) – If True, no information is logged during the optimization. Defaults to False.

Return type

Tensor

pystiche_papers.gatys_ecker_bethge_2016.preprocessor()

Return type

Module

pystiche_papers.gatys_ecker_bethge_2016.postprocessor()

Return type

Module

pystiche_papers.gatys_ecker_bethge_2016.optimizer(input_image)

Optimizer from [GEB2016].

Parameters

input_image (Tensor) – Image to be optimized.

Return type

LBFGS

pystiche_papers.gatys_ecker_bethge_2016.multi_layer_encoder(impl_params=True)

Multi-layer encoder from [GEB2016].

Parameters

impl_params (bool) – If True, the MaxPool2d in the multi_layer_encoder are exchanged for AvgPool2d.

Return type

MultiLayerEncoder

pystiche_papers.gatys_ecker_bethge_2016.compute_layer_weights(layers, multi_layer_encoder=None)

Return type

Tuple[float, ...]

pystiche_papers.gatys_ecker_bethge_2016.hyper_parameters(impl_params=True)

Hyper parameters from [GEB2016].

Return type

HyperParameters