This repository provide the code to assess the attraciveness and age of medical images. The attractiveness models are: BLINQ- trained on a private dataset obtained from the BLINQ dating app and MEDICAL- the further fine-tuned model using the Chicago Face Dataset. The age models are: WikiIMDB trained and further fine-tuned using the APPA Real Dataset. We provide the pretrained and the fine-tuned models in order to predict the attractiveness and age given a facial image.

The maximal attractiveness score is 10 and the minimal 1. However, in BLINQ the assigned scores tend to be lower than on CFD such that 10s are very rare in the former. Age is reported in the interval 0 to 100.

Python 3
pytorch>=1.3
torchvision
pillow

Please download the CNN pytorch model weight including the folders here and copy them into the checkpoints folder.

python predict_hotness.py predict_bins_hotness_densenet201_blinq path_to_img

or

python predict_hotness.py predict_bins_hotness_densenet201_medical path_to_img

python predict_age.py predict_bins_age_densenet201_wikiimdb path_to_img

or

python predict_age.py predict_bins_age_densenet201_appareal path_to_img

We run our model for demonstration purposes here on two publicly available images from the Web (downloaded from Wikipedia).

This picture of Angela Merkel achieves an attractiveness score of

2.829862910516840 with a standard deviation of 0.0877247232573163 (BLINQ) and

3.557292120774966 with a standard deviation of 0.2021892417092418 (MEDICAL).

and an age estimation of

57.4361633341181 with a standard deviation of 1.709792667712001 (WikiIMDB) and

60.83877001864456 with a standard deviation of 1.1122933985426469 (APPA Real).

Note, that we feed the image 20 times through the CNN with random augmentations to compute the mean and standard deviation.

This picture of Katheryn Winnick achieves an attractiveness score of

6.378340877147275 with a standard deviation of 0.13951626876974346 (BLINQ) and

7.932078346675053 with a standard deviation of 0.03819095674717804 (MEDICAL).

and an age estimation of

37.45339674973202 with a standard deviation of 2.8794594430390483 (WikiIMDB) and

31.53677092714849 with a standard deviation of 1.6105625494518505 (APPA Real).