restyle-encoder

ReStyle

Input

(Image from https://github.com/yuval-alaluf/restyle-encoder/blob/main/notebooks/images/)

Shape : (1, 3, 1024, 1024)

Face alignment and reshaped to : (1, 3, 256, 256)

Output

Inversion task

Toonification task

* Note: From left to right: 1st, 2nd, 3rd, 4th, 5th iteration, and original (face aligned) image.

Usage

Automatically downloads the onnx and prototxt files on the first run. It is necessary to be connected to the Internet while downloading.

For the sample image,

$ python3 restyle-encoder.py 

If you want to specify the input image, put the image path after the --input option.
You can use --savepath option to change the name of the output file to save.

$ python3 restyle-encoder.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH 

By specifying the -iter option, you can choose how many iterations you want to generate the output image (default 5).

$ python3 restyle-encoder.py -iter 3 

By specifying the -toon option, you can run the toonification task.

$ python3 restyle-encoder.py -toon --input img/toonify_img.jpg --savepath img/output_toonify.png 

By adding the --video option, you can input the video.
If you pass 0 as an argument to VIDEO_PATH, you can use the webcam input instead of the video file.

$ python3 restyle-encoder.py --video VIDEO_PATH 

By adding the --use_dlib option, you can use original version of face alignment.

Reference

• ReStyle

• PSGAN (face alignment without dlib)

Framework

Pytorch 1.10.0

Python 3.6.7+

Model Format

ONNX opset=11

Netron

restyle-encoder.onnx.prototxt

face-pool.onnx.prototxt

toonify.onnx.prototxt