The simple pytorch implementation of fast-neural-style

Paper: Perceptual Losses for Real-Time Style Transfer and Super-Resolution

train dataset : COCO 2014 Training images dataset [80K/13GB] (download).

Prerequisites

pip install requirements.txt

Usage

If you want to visualize the training process , you need to type python -m visdom.server to open visdom.

train examples:

CUDA_VISIBLE_DEVICES=3,5 python neural_style/neural_style.py train --dataset  ./data --style-image ./images/style-images/losses.jpg --save-model-dir ./my_models  --batch-size 8 --epochs 1 --cuda 1 --log-interval 5 --image-size 256

test examples:

python neural_style/neural_style.py eval --content-image ./images/content-images/r1.jpg  --model ./my_models/wave.pth --output-image wave_r1.jpg  --cuda 1

Result