Avatar-Net: Multi-scale Zero-shot Style Transfer by Feature Decoration

Unofficial Pytorch Implementation of Avatar-Net

Reference: Avatar-Net: Multi-scale Zero-shot Style Transfer by Feature Decoration, CVPR2018

Requirements

• torch (version: 1.2.0)
• torchvision (version: 0.4.0)
• Pillow (version: 6.1.0)
• matplotlib (version: 3.1.1)

Download

• The trained models can be downloaded throuth the releases.
• MSCOCO train2014 is needed to train the network.

Usage

Arguments

• --gpu-no: GPU device number (-1: cpu, 0~N: GPU)
• --train: Flag for the network training (default: False)
• --content-dir: Path of the Content image dataset for training
• --imsize: Size for resizing input images (resize shorter side of the image)
• --cropsize: Size for crop input images (crop the image into squares)
• --cencrop: Flag for crop the center reigion of the image (default: randomly crop)
• --check-point: Check point path for loading trained network
• --content: Content image path to evalute the network
• --style: Style image path to evalute the network
• --mask: Mask image path for masked stylization
• --style-strength: Content vs Style interpolation weight (1.0: style, 0.0: content, default: 1.0)
• --interpolatoin-weights: Weights for multiple style interpolation
• --patch-size: Patch size of style decorator (default: 3)
• --patch-stride: Patch stride of style decorator (default: 1)

Train example script

python main.py --train --gpu-no 0 --imsize 512 --cropsize 256 --content-dir ./coco2014/ --save-path ./trained_models/

Test example script and image

• These figures are generated in jupyter notebook. You can make the figure yourself.

Generate the stylized image with a single style (Content-style interapoltion)

python main.py --check-point ./trained_models/check_point.pth --imsize 512 --cropsize 512 --cencrop --content ./sample_images/content/blonde_girl.jpg --style ./sample_images/style/mondrian.jpg --style-strength 1.0

Generate the stylized image with multiple style

python main.py --check-point ./trained_models/check_point.pth --imsize 512 --cropsize 512 --content ./sample_images/content/blonde_girl.jpg --style ./sample_images/style/mondrian.jpg ./sample_images/style/abstraction.jpg --interpolation-weights 0.5 0.5

Generate the stylized image with multiple style and mask

python main.py --check-point ./trained_models/check_point.pth --imsize 512 --cropsize 512 --content ./sample_images/content/blonde_girl.jpg --style ./sample_images/style/mondrian.jpg ./sample_images/style/abstraction.jpg --mask ./sample_images/mask/blonde_girl_mask1.jpg ./sample_images/mask/blonde_girl_mask2.jpg --interpolation-weights 1.0 1.0

Generate the stylized image with varying patch size

python main.py --check-point ./trained_models/check_point.pth --imsize 512 --cropsize 512 --content ./sample_images/content/blonde_girl.jpg --style ./sample_images/style/mondrian.jpg --patch-size 3

Generate the stylized image with varying patch stride

python main.py --check-point ./trained_models/check_point.pth --imsize 512 --cropsize 512 --content ./sample_images/content/blonde_girl.jpg --style ./sample_images/style/mondrian.jpg --patch-stride 4