Facial Beauty Prediction

In this project, we are trying to replicate the results of this paper by estimating the attractiveness rating of faces of SCUT_FBP5500 dataset. The dataset is obtained from here.

Data cleaning is done using R to obtain the average user rating for each face in SCUT_FBP5500 dataset. These ratings can be seen in this file.

The Final Notebook is here for this project.The MAE value obtained in the paper is 0.2518 using ResNeXt-50 model while we were able to obtain 0.4967 using Resnet50 model.

The resources that helped me are:

• Imagenet with Resnet50
• Keras Applications
• Fine Tuning in Keras
• Transfer Learning & Bottlenecking