Face-Attribute-Editing-StyleGAN3

Face-Attribute-Editing-Encoder4Editing-Text2StyleGAN-InterfaceGAN-GANSpace-Based-on-StyleGAN3

Description

---

This repo is mainly to re-implement the follow face-editing papers based on stylegan

• Encoder4Editing: Designing an Encoder for StyleGAN Image Manipulation
• InterfaceGAN: InterFaceGAN: Interpreting the Disentangled Face Representation Learned by GANs
• GANSpace: GANSpace: Discovering Interpretable GAN Controls

Getting Started

Prerequisites

• Linux or macOS
• NVIDIA GPU + CUDA CuDNN
• Python 3

Installation

• Clone the repository:

git clone https://github.com/MingtaoGuo/Face-Attribute-Editing-StyleGAN3.git
cd Face-Attribute-Editing-StyleGAN3
wget https://api.ngc.nvidia.com/v2/models/nvidia/research/stylegan3/versions/1/files/stylegan3-t-ffhq-1024x1024.pkl

• Dependencies:
  We recommend running this repository using Anaconda. All dependencies for defining the environment are provided in environment.yaml.

Inference e4e

• Download the e4e pretrained model FFHQ-e4e-stylegan3

python e4e_inference.py --img_path resources/imgs/1.png --pretrain saved_models/13_12500_pspEncoder.pth

Image
Inverse

Train e4e

• Download the FFHQ dataset from NVlabs
• Download the face recognition model arcface from insightface Go to Baidu Drive -> arcface_torch -> glint360k_cosface_r50_fp16_0.1 -> backbone.pth

python e4e_train.py --data_path FFHQ --batch_size 4 --epoch 20 --arcface saved_models/backbone.pth

Inference Text2StyleGAN

python text2stylegan.py --text "a woman with blue eyes" --opt_space w --learning_rate 0.02

Description	an european woman with blue eyes	an old asian man with gray hair	an asian woman with black long straight hair	a woman with blond hair is smiling
Text2img

Inference InterfaceGAN

python interfacegan_edit.py --stylegan stylegan3-t-ffhq-1024x1024.pkl --boundary resources/interfacegan/boundary_glasses.pth

--------glasses---------	---------beard---------	---------young---------	---------smile---------

Train InterfaceGAN

We use a vision-language model CLIP as face attribute classifier to predict generated face images from StyleGAN.

python interfacegan_gendata.py --n_samples 200000 --stylegan stylegan3-t-ffhq-1024x1024.pkl 
python interfacegan_predict_attr.py --attr_text glasses
python interfacegan_linear_svm.py --path_txt resources/interfacegan/glasses_label.txt --boundary_save resources/interfacegan/boundary_glasses.pth

Inference GANSpace

python ganspace.py --stylegan stylegan3-t-ffhq-1024x1024.pkl --v_idxs 0 --layer_idxs 0-18 --n_samples 10000

v_0, layer_1_18 (gender)

v_1, layer_1_18 (rotate)

v_2, layer_1_18 (rotate + age+ bkg)

Author

Mingtao Guo E-mail: gmt798714378 at hotmail dot com

Reference

[1]. Tov, Omer, et al. "Designing an encoder for stylegan image manipulation." ACM Transactions on Graphics (TOG) 40.4 (2021): 1-14.

[2]. Shen, Yujun, et al. "Interfacegan: Interpreting the disentangled face representation learned by gans." IEEE transactions on pattern analysis and machine intelligence (2020).

[3]. Härkönen, Erik, et al. "Ganspace: Discovering interpretable gan controls." Advances in Neural Information Processing Systems 33 (2020): 9841-9850.

[4]. Radford, Alec, et al. "Learning transferable visual models from natural language supervision." International Conference on Machine Learning. PMLR, 2021.