Facial Attributes Detection, Cellebrite

Jonathan Bitton

Data Science Fellows June 2020 Cohort

Abstract

Detect facial attributes & accessories on images and label them accordingly.

Challenges (at least two)

1. Dealing with large datasets and files
2. Accuracy on small attributes
3. Trade-off between frame per second and Accuracy/Loss (choice between DL or ML models)

Achievements (according to KPIs)

1. Code library with classes such as loadmodel (for pretrained models), summarize (for metrics), preprocessing ( for data augmentation and preprocess), Train…
2. Application using Flask with inference on images

Future project development 

1. Training on more attributes
2. Generating images with unique attribute
3. Improving small attribute detection
4. Working on Neural networks to reduce the running time VS Improving performance with Machine Learning
5. Improving face detection models for not frontal images