Context and problematic

To improve product recommendation, our client wants to set up an application which is capable of analyzing a person’s skin.

Goals

The goal is to offer the possibility

Score a person’s face through a photo

Identify the determining criteria

Generate quality images of modified attributes

Our intervention

2 Data Architect, 1 Data Engineer

• Choice of different model architectures but also different preprocessing.
• Optimization Hyper Parameters.
• Production and documentation.
• Work on the interpretability of models. Visualization of internal features
• Work on image generation : how to provide information in a simple manner?
• Modify State-of-the-art architectures for our problem.
• Work on the optimization of the pipeline and the implementation of the training parallelization.
• Creation of an internal context.

Results

In-store since September 2020

Publication of a scientific article

Version 2 under development in 2021

Technical environment

Python, Tensorflow, Pytorch, Numpy, Pandas, Jupyter, Scikit-learn, OpenCV, PIL
Docker
GCP
Git