encoder4editing

encoder4editing

Input

(Image from https://github.com/omertov/encoder4editing/tree/main/notebooks/images)

Output

• inversion

  

• InterFaceGAN

  The age, smile and pose directions for the FFHQ StyleGAN Generator.

  

• GANSpace

  Editing the car domain and the face domain, taken from the the official GANSpace.

  ○ ffhq

  ○ cars

• SeFa

  Apply to the selected edit parameters.

Requirements

This model requires additional module.

pip3 install dlib     # for align face

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 encoder4editing.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 encoder4editing.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH

By adding the --model_type option, you can specify model type which is selected from "ffhq", "car", "horse", "church". (default is ffhq)

$ python3 encoder4editing.py --model_type ffhq

If the input image is aligned, specify the --aligned option.
In this case the dlib module is not needed.

$ python3 encoder4editing.py --aligned

Edit age, simile, pose by specifying --age_factor, --age_range, --smile_factor, --smile_range, --pose_factor, --pose_range parameters.

$ python3 encoder4editing.py --age_factor -3

$ python3 encoder4editing.py --age_range -5 5

The editings for the facial domain,
such as --eye_openness, --smile, --trimmed_beard, --white_hair, --lipstick options.

$ python3 encoder4editing.py --eye_openness 20 --smile -20 --trimmed_beard 20 --white_hair -24 --lipstick 20

The editings for the cars domain,
such as --car_view1, --car_view2, --car_cube, --car_color, --car_grass options.

$ python3 encoder4editing.py -i example/car_img.jpg --model_type car --car_view1 2 --car_view2 -2 --car_cube 25 --car_color -8 --car_grass -18

Specify the using index, start/end distance and step, to set the edit parameters.

$ python3 encoder4editing.py --indices 2 3 4 5 --start_distance 0.0 --end_distance 15.0 --step 3

Reference

• Designing an Encoder for StyleGAN Image Manipulation

Framework

Pytorch

Model Format

ONNX opset=12

Netron

ffhq_encoder.onnx.prototxt
cars_encoder.onnx.prototxt
horse_encoder.onnx.prototxt
church_encoder.onnx.prototxt

ffhq_decoder.onnx.prototxt
cars_decoder.onnx.prototxt
horse_decoder.onnx.prototxt
church_decoder.onnx.prototxt