Terminal Interface¶
stransfer¶
Style Transfer
stransfer [OPTIONS] COMMAND [ARGS]...
fast-st¶
Fast Style Transfer
stransfer fast-st [OPTIONS] COMMAND [ARGS]...
convert-image¶
Converts the image at image-path using the network pretrained with style-name and saves the resulting transformed image in out-dir.
A pretrained model should exist in data/models/ for the specified style-name.
The files in data/models/ contain the style names in their file names. For example, fast_st_the_great_wave_epoch1.pth was trained with the style the_great_wave
stransfer fast-st convert-image [OPTIONS] IMAGE_PATH STYLE_NAME
Options
-
-o,--out-dir<out_dir>¶ The results directory where the converted image will be saved
Arguments
-
IMAGE_PATH¶ Required argument
-
STYLE_NAME¶ Required argument
train¶
Perform the training for the fast style transfer network. A checkpoint will be created at the end of each epoch in the data/models/ directory.
stransfer fast-st train [OPTIONS] STYLE_IMAGE_PATH
Options
-
-e,--epochs<epochs>¶ How many epochs the training will take
-
-b,--batch-size<batch_size>¶ Batch size for training
-
-cw,--content-weight<content_weight>¶ The weight we will assign to the content loss during the optimization
-
-sw,--style-weight<style_weight>¶ The weight we will assign to the style loss during the optimization
Arguments
-
STYLE_IMAGE_PATH¶ Required argument
gatys-st¶
Run the original Gatys style transfer (slow). Both style-image and content-image should be the paths to the image we want to take the content from and the one we want to take the style from (respectively).
stransfer gatys-st [OPTIONS] CONTENT_IMAGE_PATH STYLE_IMAGE_PATH
Options
-
-n,--out-name<out_name>¶ The name of the result file (transformed image)
-
-s,--steps<steps>¶ How many iterations should the optimization go through.
-
-cw,--content-weight<content_weight>¶ The weight we will assign to the content loss during the optimization
-
-sw,--style-weight<style_weight>¶ The weight we will assign to the style loss during the optimization
Arguments
-
CONTENT_IMAGE_PATH¶ Required argument
-
STYLE_IMAGE_PATH¶ Required argument
video-st¶
Video Style Transfer
stransfer video-st [OPTIONS] COMMAND [ARGS]...
convert-video¶
Converts the video at video-path using the network pretrained with style-name and saves the resulting transformed video in out-dir.
A pretrained model should exist in data/models/ for the specified style-name.
The files in data/models/ contain the style names in their file names. For example, video_st_starry_night_epoch3.pth was trained with the style starry_night
stransfer video-st convert-video [OPTIONS] VIDEO_PATH STYLE_NAME
Options
-
-o,--out-dir<out_dir>¶ The results directory where the converted style will be saved
-
--fps<fps>¶ The FPS that will be used when saving the transformed video
Arguments
-
VIDEO_PATH¶ Required argument
-
STYLE_NAME¶ Required argument
train¶
Perform the training for the video style transfer network. A checkpoint will be created at the end of each epoch in the data/models/ directory.
We have the possibility of starting the training process for video style transfer from pretrained weights for a fast style transfer trained with the same style name. The weights that will be used are those of the highest epoch which correspond to that style_name.
stransfer video-st train [OPTIONS] STYLE_IMAGE_PATH
Options
-
-e,--epochs<epochs>¶ How many epochs the training will take
-
-b,--batch-size<batch_size>¶ Batch size for training
-
-cw,--content-weight<content_weight>¶ The weight we will assign to the content loss during the optimization
-
-sw,--style-weight<style_weight>¶ The weight we will assign to the style loss during the optimization
-
-tw,--temporal-weight<temporal_weight>¶ The weight we will assign to the temporal loss during the optimization
-
--use-pretrained-fast-st¶ States whether we want to start training the video model from pretrained fast style transfer weights (which was trained on the same style name)
Arguments
-
STYLE_IMAGE_PATH¶ Required argument