def test_image_optimization_optimizer_preprocessor():
input_image = torch.empty(1)
criterion = nn.Module()
optimizer = optim.default_image_optimizer(input_image)
preprocessor = nn.Module()
with pytest.raises(RuntimeError):
optim.image_optimization(input_image,
criterion,
optimizer,
preprocessor=preprocessor)
def test_default_image_optim_loop(optim_asset_loader):
asset = optim_asset_loader("default_image_optim_loop")
actual = optim.image_optimization(
asset.input.image,
asset.input.perceptual_loss,
optimizer=asset.params.get_optimizer,
num_steps=asset.params.num_steps,
quiet=True,
)
desired = asset.output.image
ptu.assert_allclose(actual, desired, rtol=1e-4)
def nst(
content_image: torch.Tensor,
style_image: torch.Tensor,
impl_params: bool = True,
hyper_parameters: Optional[HyperParameters] = None,
quiet: bool = False,
) -> torch.Tensor:
r"""NST from :cite:`GEB2016`.
Args:
content_image: Content image for the NST.
style_image: Style image for the NST.
impl_params: Switch the behavior and hyper-parameters between the reference
implementation of the original authors and what is described in the paper.
For details see :ref:`here <gatys_ecker_bethge_2016-impl_params>`.
hyper_parameters: If omitted,
:func:`~pystiche_papers.gatys_ecker_bethge_2016.hyper_parameters` is used.
quiet: If ``True``, no information is logged during the optimization. Defaults
to ``False``.
"""
if hyper_parameters is None:
hyper_parameters = _hyper_parameters()
device = content_image.device
criterion = perceptual_loss(impl_params=impl_params,
hyper_parameters=hyper_parameters)
criterion = criterion.to(device)
input_image = misc.get_input_image(
starting_point=hyper_parameters.nst.starting_point,
content_image=content_image)
preprocessor = _preprocessor().to(device)
postprocessor = _postprocessor().to(device)
criterion.set_content_image(preprocessor(content_image))
criterion.set_style_image(preprocessor(style_image))
torch.autograd.set_detect_anomaly(True)
return optim.image_optimization(
input_image,
criterion,
optimizer=optimizer,
num_steps=hyper_parameters.nst.num_steps,
preprocessor=preprocessor,
postprocessor=postprocessor,
quiet=quiet,
)
def test_default_image_optim_loop_processing(optim_asset_loader):
asset = optim_asset_loader("default_image_optim_loop_processing")
actual = optim.image_optimization(
asset.input.image,
asset.input.criterion,
optimizer=asset.params.get_optimizer,
num_steps=asset.params.num_steps,
preprocessor=asset.params.preprocessor,
postprocessor=asset.params.postprocessor,
quiet=True,
)
desired = asset.output.image
ptu.assert_allclose(actual, desired, rtol=1e-4)
def main(raw_args: Optional[List[str]] = None) -> None:
args = parse_args(raw_args)
config = make_config(args)
config.perceptual_loss.set_content_image(config.content_image)
config.perceptual_loss.set_style_image(config.style_image)
output_image = image_optimization(config.input_image,
config.perceptual_loss,
num_steps=config.num_steps)
if not config.output_image_specified:
print(f"Saving result to {config.output_image}")
write_image(output_image, config.output_image)
# .. code-block:: python # # starting_point = "random" # input_image = get_input_image(starting_point, content_image=content_image) ######################################################################################## # Finally we run the NST with the :func:`~pystiche.optim.image_optimization` for # ``num_steps=500`` steps. # # In every step the ``perceptual_loss`` is calculated nd propagated backward to the # pixels of the ``input_image``. If ``get_optimizer`` is not specified, as is the case # here, the :func:`~pystiche.optim.default_image_optimizer`, i.e. # :class:`~torch.optim.LBFGS` is used. output_image = optim.image_optimization(input_image, perceptual_loss, num_steps=500) ######################################################################################## # After the NST is complete we show the result. # sphinx_gallery_thumbnail_number = 4 show_image(output_image, title="Output image") ######################################################################################## # Conclusion # ---------- # # If you started with the basic NST example without ``pystiche`` this example hopefully # convinced you that ``pystiche`` is a helpful tool. But this was just the beginning:
criterion = loss.PerceptualLoss(content_loss, style_loss).to(device) print(criterion) ######################################################################################## # We set the target images for the optimization ``criterion``. criterion.set_content_image(content_image) criterion.set_style_image(style_image) ######################################################################################## # We perform the unguided NST and show the result. starting_point = "content" input_image = get_input_image(starting_point, content_image=content_image) output_image = optim.image_optimization(input_image, criterion, num_steps=500) ######################################################################################## show_image(output_image) ######################################################################################## # While the result is not completely unreasonable, the building has a strong blueish # cast that looks unnatural. Since the optimization was unconstrained the color of the # sky was used for the building. In the remainder of this example we will solve this by # dividing the images in multiple separate regions. ######################################################################################## # Guided image optimization # ------------------------- #
