Overview / Usage

This Deep Learning pipeline is capable of detecting and localizing faces in an input image. This is achieved through Multi Task Cascaded CNN architecture. This architecture is implemented in TensorFlow. The later stage of this pipeline include a multi task classifier architecture capable of determining age, ethnicity, gender and emotion of the detected face. The architecture is trained using UTKFace dataset.

Methodology / Approach

The following methodology was used to build this pipeline:

1. Collected dataset for training purpose.
2. Looked for DL architectures capable of detecting faces.
3. Looked for an implementation of that architecture and the pretrained weights if any.
4. Once the detection part was up and running, EDS of UTKFace dataset began.
5. Important insights were drawn form the dataset and architecture solution was devised accordingly.
6. Build various architectures to experiment with.
7. Settled on one which worked.
8. Implemented the pipeline.

Technologies Used

1. TensorFlow
2. Keras
3. OpenCV

Repository

https://github.com/ayulockin/humAIn_FaceLess