https://github.com/adityajn105/automatic-kinship-detection

This is a kaggle Challenge. Given a pair of images of 2 faces we have to determine whether they are related or not. Here I have used a Siamese network over VGG-facenet to tackle this problem.
https://github.com/adityajn105/automatic-kinship-detection

facenet-model kaggle-competition keras kinship siamese-network

Last synced: 10 days ago
JSON representation

This is a kaggle Challenge. Given a pair of images of 2 faces we have to determine whether they are related or not. Here I have used a Siamese network over VGG-facenet to tackle this problem.

• Host: GitHub
• URL: https://github.com/adityajn105/automatic-kinship-detection
• Owner: adityajn105
• License: gpl-3.0
• Created: 2019-05-25T08:00:47.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2019-05-29T04:07:55.000Z (over 5 years ago)
• Last Synced: 2024-11-15T14:25:09.900Z (2 months ago)
• Topics: facenet-model, kaggle-competition, keras, kinship, siamese-network
• Language: Jupyter Notebook
• Homepage: https://ieeexplore.ieee.org/document/8756528
• Size: 847 KB
• Stars: 1
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Automatic-kinship-detection

This is a kaggle Challenge. Given a pair of images of 2 faces we have to determine whether they are related or not. Here I have used a Siamese network over VGG-facenet to tackle this problem.

[Fork the solution notebook in Colab](https://colab.research.google.com/drive/1sPaawdP75srFpI2sVi3RqJzp1F4KgtYj)

## Problem

To build a more complex model by determining if two people are blood-related based solely on images of their faces. Not every individual in a family shares a kinship relationship. For example, a mother and father are kin to their children, but not to each other. Below is the example of kinship where person in a specific row are related.

![Example Faces](screenshots/kinship_detection.gif)

## Solution

Here I have used a siamese architecture which is built using VGG-Face resnet. Loss used is binary_crossentropy. Optimizer is Adam. My latest submission scored around 0.792. Training and Validaiton metrics of Trained model are given below.

![Performance](screenshots/performance.png)

## Architecture Diagram

![Architecture](screenshots/kinship_arch.png)

## Authors

* Aditya Jain : [Portfolio](https://adityajain.me)

## Licence

This project is licensed under the MIT License - see the [LICENSE.md](https://github.com/adityajn105/Automatic-kinship-detection/blob/master/LICENSE) file for details

## Must Read

1. [Triplet Loss](https://omoindrot.github.io/triplet-loss)