A Unified Approach to Kinship Verification

TL;DR

This paper introduces a deep learning framework for kinship verification that employs multi-task learning, novel embedding fusion, and adaptive sampling to improve accuracy on small, imbalanced datasets, outperforming current methods.

Contribution

It presents a unified multi-task deep learning approach with innovative embedding fusion and adaptive sampling for kin verification, addressing dataset imbalance and overfitting.

Findings

Outperforms state-of-the-art kin verification methods

Effective on multiple datasets including Families In the Wild, FG2018, FG2020

Demonstrates robustness with small and imbalanced datasets

Abstract

In this work, we propose a deep learning-based approach for kin verification using a unified multi-task learning scheme where all kinship classes are jointly learned. This allows us to better utilize small training sets that are typical of kin verification. We introduce a novel approach for fusing the embeddings of kin images, to avoid overfitting, which is a common issue in training such networks. An adaptive sampling scheme is derived for the training set images to resolve the inherent imbalance in kin verification datasets. A thorough ablation study exemplifies the effectivity of our approach, which is experimentally shown to outperform contemporary state-of-the-art kin verification results when applied to the Families In the Wild, FG2018, and FG2020 datasets.