Transfer learning induced DeIT for alcohol consumption detection

A Pytorch C++ based API for alcohol consumption detection using periocular NIR images on mobile devices. Here, we employ transfer learning induced Data-efficient Image Transformer(DeIT) model for alcohol consumption detection using periocular NIR iris images dataset.

Deployment Code:

Follow these steps to utilize the API for your mobile devices:

1. git clone https://github.com/system-reboot/DeiT-for-alcohol-consumption-Detection.git
2. cd DeiT-for-alcohol-consumption-Detection
3. mkdir build/ && cd build/
4. cmake -DCMAKE_PREFIX_PATH=<absolute_path_to_libtorch>
5. make
6. ./Alcohol-Consumption-Detector <path_to_your_image_file>

Files:

1. model/binary_class_model.ipynb - Model trained to detect if the subject is under alcohol consumption or not.
2. model/multi_class_model.ipynb - Model trained to study the temporal impact of alcohol on iris central nervous system(CNS).
3. src/main.cpp - Sample main file to load and preprocess sample images which is then classified using the pre-trained model.

Note:

The weights utilized for training the above model has not published due to privacy concerns of the subjects in the dataset. One can contact Juan Tapia Farias regarding the datset.