Demo codes for

"Face Detection in Untrained Deep Neural Networks"

Seungdae Baek, Min Song, Jaeson Jang, Gwangsu Kim, and Se-Bum Paik*

*Contact: sbpaik@kaist.ac.kr

1. System requirements

• MATLAB 2019b or later version
• Installation of the Deep Learning Toolbox (https://www.mathworks.com/products/deep-learning.html)
• Installation of the pretrained AlexNet (https://de.mathworks.com/matlabcentral/fileexchange/59133-deep-learning-toolbox-model-for-alexnet-network)
• No non-standard hardware is required.
• Uploaded codes were tested using MATLAB 2019b.

2. Installation

• Download all files and folders. ("Clone or download" -> "Download ZIP")
• Download 'Data.zip', 'Stimulus.zip' from below link and unzip files in the same directory
• [Data URL] : https://doi.org/10.5281/zenodo.5637812
• Expected Installation time is about 45 minutes, but may vary by system conditions.

3. Instructions for demo

• Run "Main.m" and select result numbers (from 1 to 6) that you want to perform a demo simulation.
• Expected running time is about 5 minutes for each figure, but may vary by system conditions.

4. Expected output for demo

• Below results for untrained AlexNet will be shown.
  • Result 1) Run_Unit: Spontaneous emergence of face-selectivity in untrained networks (Fig.1, Fig.S1-3)
  • Result 2) Run_PFI: Preferred feature images of face-selective units in untrained networks (Fig.2, Fig.S4)
  • Result 3) Run_SVM: Detection of face images using the response of face units in untrained networks (Fig.3, Fig.S11-12)
  • Result 4) Run_Trained: Effect of training on face-selectivity in untrained networks (Fig.4)
  • Result 5) Run_Invariance: Invariant characteristics of face-selective units in untrained networks (Fig.S5)
  • Result 6) Run_View: Viewpoint invariance of face-selective units in untrained networks (Fig.S8)