Facial Attribute Recognition from face images

This is a Keras implementation of a CNN for facial attribute recognition. I trained Visual Transformer and facenet for facial attribute extraction.

Dependencies

• Python3.6+

Tested on

• Ubuntu 16.04, Python 3.6.9, Tensorflow 2.3.0, CUDA 10.01, cuDNN 7.6

Dataset

I trained the face attribute extraction models with CelebFaces Attributes (CelebA) Dataset

You can download the preprocessed dataset from the below link. I cropped the faces and converted them into RGB format. The dataset contains 100000 images with facial attributes. https://drive.google.com/drive/folders/1iffYL-rB-3MbqI-TfFFHU6Wc-JaYHgGz?usp=sharing

Train

python train.py --imagepath=/data/imageFile100000.npz --labelpath=/data/labelFile100000.npz

Testing

python demo.py

Pre_trained weights

Please use the below weights for testing. https://drive.google.com/drive/folders/1NWz2E3b75mO_Ox8tb9d77vBi8dNHUv1T?usp=sharing

Model results:

Validation Dataset results

Test sample

I used the bigbangtheory cast image as a testing image. Please find the person’s result.

Output Video

References:

FaceNet Tensorflow

Vision Transformer (ViT)

CelebFaces Attributes (CelebA) Dataset

GitHub

View Github