An attempt to develop a toolkit for running and evaluating privacy preserving techniques in facial biometrics

1. Features
2. Prerequisites
3. Installation
4. Configuration
5. Usage
6. Examples
7. Project Structure
8. Contributing
9. License
10. Contact

--

• Command-line interface (something like https://docs.python.org/3.8/library/cmd.html):
  • easy to use (simple commands + names to run experiments)
  • easy to generate results
  • good look & feel
  • responsive logging (showing % of performed actions)
  • helpful tips (where results are saved, how can be visualised, etc)
• Handling of multiple separated virtual environments
  • running the models in subshells, while reporting progress to main interface
• Configuration parameters in text format via .ini file (loading & saving of different configurations within command line)
• Checking and handling of datasets
  • Raw datasets path (simple names)
  • Cropped & Aligned path (dataset preparation/standardization)
• Saving intermediate results for each phase (for each dataset, for each of the models)
• Saving final results (final deidentified images and dataset evaluation scores / plots)
• Handling and storing models / binary files for existing techniques

*   Pretrained models directory 

--

• Operating System: Linux ?
• Python: 3.9+
• Additional Dependencies: Conda, Mamba

1. Clone the project:

   git clone https://github.com/blazm/deid-toolkit

2. Get techniques.zip and aligned.zip (and original.zip if wanted) and extract them with unzip:

   unzip techniques.zip -d root_dir
   unzip evaluation.zip -d root_dir
   unzip visualization.zip -d root_dir
   unzip aligned.zip -d root_dir/datasets
   unzip original.zip -d root_dir/datasets

3. Create the toolkit environment:

   conda env create -f toolkit.yml

4. In deid_shell.py, change the conda_sh_path constant with the correct path to the conda.sh file on your machine.

• root - see currently set root directory
• set root - set root directory
• serve - run webserver to see the generated results
• set serve - set results directory for serving results
• load config "filename.ini"
• save config "filename.ini"
• help "command"
• ? - list of all commands

• datasets
• techniques
• evaluation
• visuals
• selection

• select datasets
• select techniques
• select evaluation
• select *

• run preprocess
• run techniques
• run evaluation
• run visualize
• run *

--

This module manages the datasets required for de-identification. It’s the first part of the pipeline. The toolkit is able to integrate (<-----) additional facial images datasets. Moreover, the datasets won’t be included in the toolkit because some of them have different licensing constraints.