Face Masker & Face Morphing: Research Stimulus Automation

A collection of AI-enhanced tools pertaining to an ongoing research on the impact of different Carotene levels on the perception of attractiveness and health.

The repo consists of:

• A Face Morphing tool: A tool designed to provide latent-based morphs of individuals providing a means to anonymize participants in research studies. The tool utilizes a method called Pivotal Tuning Inversion to create near perfect recreations of participant's faces in W+ space; enabling non-linear interpolation between three particpants at a time.
• A Face Masking tool: Designed to apply progressively stronger tints to targeted areas of an image to simulate various shades along the CIELAB colour space in accordance with established methodologies. The tool utilizes a pretrained Facial Segmentation model to accurately mask skin areas while preserving natural features such as the eyes, lips, and hair.

Installation and Usage

Face Morphing

1. Create a copy of the kaggle notebook from here
2. Enable Internet under the Settings tab
3. Select GPU P100 as the GPU ACCELERATOR option under the settings tab
4. Upload a zip file with a nested folders containing three images each. The names of these files are unimportant. An example of this is seen below:

Zip File
    > Morph_1_Folder
        > Person1.png
        > Person2.png
        > Person3.png
    > Morph_2_Folder
        > ...

5. Start PTI
6. The notebook WILL crash on the first run. Run the second code cell again and it will function correctly
7. Download the results.

Face Masking

1. Git clone this repository

git clone https://github.com/RubinInsert/FaceMasker

2. OR, Download it manually from github
3. Move into the directory with the python code On Windows, this would be:

cd <Folder Directory>

5. Install the requirements

pip install -r requirements.txt

4. Ensure your folder structure has the following format. Not all folders are required to run the program. However, any morph images must be found at the end of a correct folder trail like Input Folder > Male > Caucasian > 18-24 > M1.png and must be named correctly as M1.png-M10.png:

Input Folder
    > Male
        > Caucasian
            > 18-24
            > 25-34
            > 35-44
            > 45-54
            > 55-64
            > 65 and over
        > African
            > ...
        > Asian
            > ...
    > Female
        > ...

5. Run the program using

python3 main.py -input "Example_Folder_Input/" -output "Example_Folder_Output/" -color_tint 2.36,0,0 -thread_count 2

input dictates the input folder

output dictates the output folder

color_tint dictates what colour to apply to masked faces at each step. If more than one value is zero, the processing time will be longer.

thread_count dictates how much processing is allocated to the program. 2-4 is generally a reasonable value.

Results

Face Morphing

Face Masking

Resources

• NVIDIA's Stylegan2-ada-pytorch model
• Daniel Roich's Pivotal Tuning for Latent-based editing of Real Images
• zllrunning's Facial Parsing Map implementation extended from a BiSeNet backbone trained on the CelebAMask-HQ dataset
• Kaggle Notebook