Command line

stylish

Style transfer using deep neural network

stylish [OPTIONS] COMMAND [ARGS]...

Options

--version

Show the version and exit.

-v, --verbosity <verbosity>

Set the logging output verbosity. [default: info]

Options:debug|info|warning|error

apply

Apply a style generator model to an image.

stylish apply [OPTIONS]

Options

--model <model>

Path to trained style generator model which will be used to apply the style. [required]

-i, --input <input>

Path to image to transform. [required]

-o, --output <output>

Path to folder in which the transformed image will be saved. [required]

download

Download necessary elements to train a style generator model

Example:

 stylish download vgg19 stylish download coco2014 -o /tmp
stylish download [OPTIONS] COMMAND1 [ARGS]... [COMMAND2 [ARGS]...]...

coco2014

Download COCO 2014 Training dataset (13GB).

stylish download coco2014 [OPTIONS]

Options

-o, --output <output>

Output path to save the element (Current directory is used by default)

vgg19

Download pre-trained Vgg19 model (549MB).

stylish download vgg19 [OPTIONS]

Options

-o, --output <output>

Output path to save the element (Current directory is used by default)

train

Train a style generator model.

stylish train [OPTIONS]

Options

--vgg <vgg>

Path to Vgg19 pre-trained model in the MatConvNet data format.

-s, --style <style>

Path to image from which the style features will be extracted.

-t, --training <training>

Path to a training dataset folder (e.g. COCO 2014).

--limit <limit>

Maximum number of files to use from the training dataset folder.

-l, --learning-rate <learning_rate>

Learning rate for optimizer. [default: 0.001]

-b, --batch-size <batch_size>

Batch size for training. [default: 4]

-e, --epochs <epochs>

Epochs to train for. [default: 2]

-C, --content-weight <content_weight>

Weight of content in loss function. [default: 7.5]

-S, --style-weight <style_weight>

Weight of style in loss function. [default: 100.0]

-T, --tv-weight <tv_weight>

Weight of total variation term in loss function. [default: 200.0]

-L, --layer-weights <layer_weights>

Weights of layers used for style features extraction. [default: 1.0, 1.0, 1.0, 1.0, 1.0]

-o, --output <output>

Path to folder in which the trained model will be saved.