Facial Beauty Prediction
Personal Project @ CSIE 5420 Cognitive Computing
Personal Project @ CSIE 5420
Cognitive Computing
A framework to realtime evaluate facial beauty.
Figure 1. Evaluated example with different saturation and value (HSV color).
Facial beauty prediction is a significant visual recognition problem to make assessment of facial attractiveness that is consistent to human perception. In this project, I plan to implement a framework to learn the distribution of facial beauty and help machine develop cognition about attractiveness. Moreover, I discuss the effectiveness for future applications regarding different skin colors, lightness, and saturation of images. In this work, the beauty scores are range from 1-5 and my results are shown above in Figure 1.
Figure 2. Comparison of different ImageNet models and objective functions by 5-fold cross validation.
As for the proposed method, I used different pre-trained ImageNet models to verify the capabilities. Also, I used different objective functions to deal with the class relationship problem which we cannot view the beauty scores as independent class, including Earth Mover Distance-based loss, Mean Square Error, and Multi-Binary Cross Entropy Loss. I evaluated my results with pearson correlation (PC), maximum absolute error (MAE), and root mean square error (RMSE) which is listed above in Figure 2. For more information, please refer to my slides and my code is also publically available on GitHub.