kwjinwoo/real_time_style_transfer

Implementation Perceptual Losees for Real-Time Style Transfer and Super-Resolution with TF2

Real_time_style_transfer

Introduction

• This repository is implementation of Perceptual Losses for Real-Time Style Transfer and Super-Resolution
• You can do real time style transfer
• Difference from origin paper
  • for upsampling, use Conv2DTranspose with padding
  • for style, content loss, use squared sum instead norm

Requirement

python == 3.8
tensorflow == 2.8.0
opencv-python == 4.5.5

Dataset

• use COCO2014 Validation image dataset
• Dataset structure

  $root/
  ├── val2014
  

Train

• Run

  python train.py --style_img --dataset_dir
  

• Args
  • style_img : target style image path
  • dataset_dir : dataset dir path
• Returns
  • ckpt : model checkpoint at every 100 iter. saved at "./ckpt" dir
  • model : final model. saved at "./models" dir

Inference

• Run

  python infer.py --test_dir --model_path
  

• Arags
  • test_dir : test images dir path that you want style transfer
  • model_path : saved model path
• Returns
  • result : transfered images saved at "./reslut" dir

Result

• Content image
  
• Transfer result
  

Reference

[1] Perceptual Losses for Real-Time Style Transfer and Super-Resolution. Justin Johnson, Alexandre Alahi, Li Fei-Fei