Looking for facial beauty prediction dataset

I have come across this article: https://towardsdatascience.com/how-attractive-are-you-in-the-eyes-of-deep-neural-network-3d71c0755ccc and this paper https://arxiv.org/pdf/1801.06345.pdf earlier and was inspired to build a facial beauty classifier.

However, my model seems to overfit on the training set and produces unstable prediction on new data. The dataset in the paper only contains 5500 labeled images.

I have been looking for new training data.

Another dataset mentioned in the paper.

(here is the mention and papers referenced)

" Hot-Or-Not database [32][33]"

[32] R. White, A. Eden, and M. Maire, “Automatic prediction of human
attractiveness,” UC Berkeley CS280A Proj., vol. 1, p. 2, 2004.
[33] D. Gray, K. Yu, W. Xu and Y. Gong, “Predicting facial beauty without
landmarks,” in Proc. ECCV, 2010.

I have been looking for this HotOrNot dataset and googling these two papers but found nothing.
Hot or Not seems to switch to a Like or Dislike system from the 0 - 10 score system that came before.

I am wondering if anyone has any lead on this “Hot or Not dataset”? (or any kind of dataset that has human - face labeled with facial beauty score)

What one person finds beautiful, another person might not.
The model you train in with any dataset will just replicate the taste of the person who created the dataset.

You ca try this:

  • Perform unsupervised clustering of images based on the facial features.
  • Make the user classify a few images from each cluster.
  • When presented with a new image, the model should allocate it to a cluster and use the predictions made by the user in previous step to predict Beautiful or not.

This will replicate the user’s taste.

A bit late, but you can find the dataset you were looking for here.

1 Like