boscong88 / beauty.torch

Understanding facial beauty with deep learning.

beauty.torch

This project serves a deep learning model scoring selfie images between 1 to 10 based on image and face attributes. You can learn the technical details of this project from this blog post. Use [resnet.torch] (https://github.com/erogol/resnet.torch), if you plan to follow all the training pipeline described on the post.

Given image is processed as follows;

1. Detect face.
2. Find landmarks
3. Rotate image to align face.
4. Fill gaps with constant pixel value.
5. Send into scoring model.

For an example use check notebook ExampleUse.ipynb

Requirements

Main requirement is [Torch](http://torch.ch/docs/getting-started.html) computing framework.

Models

[dlib face model](http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2) - place under ```utils/``` [beauty model](https://www.dropbox.com/s/yezp73uxqnd86e6/model_best.t7) - GPU model (use ```utils/convert2cpu.lua``` for setting it for CPU) place under ```trained/``` [optimstate](https://www.dropbox.com/s/yezp73uxqnd86e6/model_best.t7) - if you like to fine-tune the model.

Python

dlib ```sudo pip install dlib``` - face and landmark detection) lutorpy ```sudo pip install lutorpy``` - using torch model on python skimage ```sudo pip install skimage``` - image processing cv2 ```sudo pip install cv2``` - OpenCV python module

What you have here useful

* Face alignment code in ```utils/img_processing.py```. * A template for porting Torch models to python in ```utils/Classifier.py```. * The model itself

Examples

Attention of the trained model. 

Sorting A. Lima images from Google Search.