tanelp / neural-obfuscator

neural-obfuscator

A library for anonymizing identities in an image by swapping faces to the ones that have never existed before.

Installation

Installing from source

To install neural-obfuscator from the GitHub source, download the project:

git clone https://github.com/tanelp/neural-obfuscator.git

Then, move to the folder and run the install command:

cd neural-obfuscator
pip install --editable .

Usage

Pixelating a face

import cv2
import neural_obfuscator as no

model = no.Obfuscator(method="pixelate")
img = cv2.imread("path/to/image.jpg")
img_gdpr = model.obfuscate(img)
no.show_image(img_gdpr)

Swapping to a random face

import cv2
import neural_obfuscator as no

model = no.Obfuscator(method="swap")
img = cv2.imread("path/to/image.jpg")
img_gdpr = model.obfuscate(img)
no.show_image(img_gdpr)

Synthesysing a face with a StyleGAN model

The repository includes a PyTorch implementation of StyleGAN.

import neural_obfuscator as no

model = no.StyleGAN(weights="ffhq")
latents = np.random.RandomState(5).randn(1, 512).astype(np.float32)
latents = torch.from_numpy(latents)
imgs = model.forward(latents)

no.show_image(imgs[0])

Encoding a face into StyleGAN's latent space

landmarks = landmarks_model.predict(img, face)
img_aligned, _ = aligner.align(img, landmarks, method="eyes_nose")
dlatents = encoder.encode(img_aligned, optim_image_size=256, num_steps=300)

Check examples/encode.py for a full example.