This repository contains the scripts for training simple glasses classifier.

Create new environment via conda and inside it run:

$ pip install -r requirements.txt

The well-known CelebA dataset (in-the-wild images) was used to train and test the model.

You need to run the following script to preprocess images obtained in the wild to train a simple classifier.

$ python scripts/extract_eyes_region.py --img_folder=dataset/img_celeba --output_folder=cropped_images
$ python scripts/extract_eyes_region.py --help #For more information about parameters

$ python train.py

$ python scripts/inference_on_images.py
$ python scripts/inference_on_images.py --help #For more information about parameters
$ python scripts/inference_on_images.py --time #If you want check inference time

There are two trained models in this repository.

• Simple small VGG-like (it classifies cropped image which obtains via landmark detector)

• MobileNetV3 (it classifies whole image)

For MobileNet-v3 you can look at MobileNetV3/mobile-net.ipynb