Project 5. The similarities and differences between deep convolutional neural networks (DCNN) and human behavior in face perception

Design background

• A very classic phenomenon in face processing is the face inversion effect, which refers to the fact that people exhibit a greater advantage in processing upright faces compared to inverted faces.
• Research indicates that the face inversion effect is inherent or innate.
• Deep convolutional neural networks (DCNN) can also recognize human faces after training. Whether DCNN with different learning experiences can exhibit human-like faces inversion effects?

fig5-1

Methods

1. We trained AlexNet to classify the orientation of face pictures from VGG face dataset.
2. We also trained another AlexNet to classify the orientation of many different categorys pictures from ImageNet dataset.

fig5-2

Results

• We got basic results.
• Unpublished results, for more information, please contact me.

fig5-3

Co-author: Xi Wang, Ms., Xiqian Lu, Dr. & Yi Jiang, Prof.
Copyright: Ruidi Wang.
Contact: [email protected] & [email protected]