pystiche_papers.ulyanov_et_al_2016.rst

pystiche_papers.ulyanov_et_al_2016

Title	Texture Networks: Feed-forward Synthesis of Textures and Stylized Images
Authors	Dmitry Ulyanov, Vadim Lebedev, Andrea Vedaldi, and Viktor S. Lempitsky
Citation	:cite:`ULVL2016` / :cite:`UVL2017`
Reference implementation	Repository / Archive
Variant	Model optimization
Content loss	:class:`~pystiche.loss.FeatureReconstructionLoss`
Style loss	:class:`~pystiche.loss.GramLoss`

Instance norm

The authors published an improved version :cite:`UVL2017` of their initial paper :cite:`ULVL2016` with only a single but significant change: they developed :class:`~torch.nn.InstanceNorm2d` and used it as drop-in replacement for :class:`~torch.nn.BatchNorm2d` layers. To account for this we provide instance_norm flag, which defaults to True.

The original authors also use the same repository for both implementations and only differentiate between them with the branches:

Paper	Branch
:cite:`ULVL2016`	texture_nets_v1
:cite:`UVL2017`	master

Behavioral changes

.. seealso::
  :ref:`Paper implementations <impl_params>`

The following parts are affected:

• :func:`~pystiche_papers.ulyanov_et_al_2016.content_transform`,
• :func:`~pystiche_papers.ulyanov_et_al_2016.GramLoss`,
• :func:`~pystiche_papers.ulyanov_et_al_2016.ConvBlock`,
• :func:`~pystiche_papers.ulyanov_et_al_2016.level`,
• :func:`~pystiche_papers.ulyanov_et_al_2016.Transformer`.

Hyper parameters

.. seealso::
  :ref:`Paper implementations <impl_params>`

Although there are four possible combinations for impl_params and instance_norm only three are listed below. Since the hyper-parameters in both papers are equal, both combinations with ìmpl_params=False are equal and thus not reported.

:func:`~pystiche_papers.ulyanov_et_al_2016.content_loss`

Parameter	impl_params / ìnstance_norm
	True / True	True / False	False
layer	"relu4_2"
score_weight	1e0	6e-1	1e0

:func:`~pystiche_papers.ulyanov_et_al_2016.style_loss`

Parameter	impl_params
	True / True	True / False	False
layers	("relu1_1", "relu2_1", "relu3_1", "relu4_1")	("relu1_1", "relu2_1", "relu3_1", "relu4_1", "relu5_1")
layer_weights	"sum"
score_weight	1e0	1e3	1e0

:func:`~pystiche_papers.ulyanov_et_al_2016.content_transform`

Parameter	Value
edge_size	256

:func:`~pystiche_papers.ulyanov_et_al_2016.style_transform`

Parameter	impl_params / ìnstance_norm
	True / True	True / False	False
edge_size	256
edge	"long"
interpolation_mode	"bicubic"	"bilinear"

:func:`~pystiche_papers.ulyanov_et_al_2016.batch_sampler`

Parameter	impl_params / ìnstance_norm
	True / True	True / False	False
num_batches	2_000	300	200
batch_size	1	4	16

:func:`~pystiche_papers.ulyanov_et_al_2016.optimizer`

Parameter	impl_params / ìnstance_norm
	True / True	True / False	False
lr	1e-3	1e0

:func:`~pystiche_papers.ulyanov_et_al_2016.lr_scheduler`

Parameter	impl_params / ìnstance_norm
	True / True	True / False	False
lr_decay	0.8		0.7
delay	0		4

Miscellaneous

Parameter	impl_params / ìnstance_norm
	True / True	True / False	False
num_epochs	25	10

API

.. automodule:: pystiche_papers.ulyanov_et_al_2016

.. autofunction:: content_transform

.. autofunction:: style_transform

.. autofunction:: images

.. autofunction:: dataset

.. autofunction:: image_loader

.. autofunction:: content_loss

.. autoclass:: GramLoss

.. autofunction:: style_loss

.. autofunction:: perceptual_loss

.. autofunction:: transformer

.. autofunction:: training

.. autofunction:: stylization

.. autofunction:: hyper_parameters

.. autofunction:: multi_layer_encoder

.. autofunction:: preprocessor

.. autofunction:: postprocessor

.. autofunction:: optimizer

.. autofunction:: lr_scheduler