Identification of Kinship Relationships
- 0 Collaborators
In most of the cases kin relationship means Blood relationship is identified from genes of father/mother - child, which involves somewhat Biological terms, But now we want to make machine or model that can predict kin relationship using Computer Vision - one of the subfield of Deep Learning . ...learn more
Project status: Concept
Intel Technologies
Intel Integrated Graphics,
Intel Opt ML/DL Framework,
Intel CPU,
Intel Python
Overview / Usage
This is basically Kaggle Competition and have world wide participation of different learners. Competition is having around 1,00,000 to 2,00,000 different face photo images - Somewhat blurry, zoomed, rotated which is divided into different folders as per family. So first tradious task to make dataset which can be done by using Probability Sampling and Data Generator. Then there is so many work based experience is done during this project like dealing with libraries, data structures, different types of model structure, setting up appropriate tuning and hyper parameter etc. Then finally it solves classification with state of art performance, where humans also find difficult to predict kinship relationship by seeing only Face photos.
Methodology / Approach
Project is having Deep Learning model that can predict whether two person are Blood related or not by inputting their Face photos. Model uses Transfer learning with pre trained models to make converge rate faster.. Now to reduce average error of final model Ensemble Learning technique is used.
Technologies Used
-
Pandas
-
Numpy
-
Tensorflow
-
Keras
-
os, glob