fast-neural-style-super-resolution

implement fast-neural-style-super-resolution by pytorch

In this repository we only provide the source code.

Setup

All code is implemented in PyTorch.

See more details in PyTorch official website.

Training a new model

Use neural_style_vx_x.py to train a new model. Use train key word to set the script in train mode.

Model options for training:

• --epoch: How many epoch would you like to do. The default is 2.

• --save-model-dir: The directory that the model will be saved.

• --cuda: Set 1 to use GPU

• --arch: The architecture of the transform net. It isn't implemented yet.

• --batch-size: Default is 4

• --dataset: The dataset you used to train the model. The floder must contain another folder which contain all images.

• --upsample: The scale of the edge.

• --image-size: All images will be resize in the image_size

• --seed: Random seed for training

• --content-weight: Weight for content-loss, default is 1.0

• --pix_weight: Weight for pixel-loss, default is 1.0

• --lr: Learning rate. Default is 0.001

• --log-interval: Number of images after which the training loss is logged, default is 500

• srcnn: The architecture of the transform net. set 1 to use srcnn. It isn't implemented yet.

Use eval to super resolution a image.

Modle options for super resolution

• --content-image: The directory of the LR image.

• --content-scale: Factor for scaling down the content image.

• --output-image: Path for saving output image.

• --model: Saved model to be used for super resolution.

• --cuda: Set it to 1 for running on GPU, 0 for CPU