our paper "Adaptively Weighted k-tuple Metric Network for Kinship Verification" has been accepted by IEEE Transactions on Cybernetics (Impact Factor=11.448)
KINSHIP-VERIFICATION
Inspired by human visual systems that incorporate both low-order and high-order cross pair information from local and global perspectives, we propose to leverage high-order cross-pair features and develop a novel end-to-end deep learning model named the Adaptively Weighted k-Tuple Metric Network (AWk-TMN).
First, a novel cross-pair metric learning loss based on k-tuplet loss is introduced. It naturally captures both the low-order and high-order discriminative features from multiple negative pairs. Second, an adaptively weighting scheme is formulated to better highlight hard negative examples among multiple negative pairs, leading to enhanced performance. Third, different levels of convolutional features are integrated to further exploit the low-order and high-order representational power, with jointly optimized feature and metric learning. Extensive experimental results on three popular kinship verification datasets demonstrate the effectiveness of the proposed AWk-TMN approach when comparing with the state-of-the-art approaches.
conda create --n kinship python=3.7
conda activate kinship
# install dependencies
pip install -r requirements.txt
# install torch and torchvision (select the proper cuda version to suit your machine)
conda install pytorch torchvision cudatoolkit=11.1 -c pytorch
- pip install -r requirements.txt
- modify config file(configs/configs.yml) to point to the data set
- type the following code
python kinship.py --conv=conv1234 --dataset=KINFACE1
Adaptively Weighted k-tuple Metric Network for Kinship Verification
python eval/kinship_eval.py --conv conv1234 -t KINFACE1 --feature_encoder './model/KINFACE1_model/feature_encoder_KINFACE1_KINFACE1_conv1234_awk_K_PAIR_2_SPLIT_1_a_0.6_m_0.4.pkl' --relation_network './model/KINFACE1_model/relation_network_KINFACE1_KINFACE1_conv1234_awk_K_PAIR_2_SPLIT_1_a_0.6_m_0.4.pkl'
python test/kinship_test.py --conv=conv1234 --feature_encoder='yourmodelpath' --relation_network='yourmodelpath' --img1='the first img path of the pair-imgs you wanna test' --img2='the second img path of the pair-imgs you wanna test'
kinship.py
you can train our network with this fileeval/kinship_eval.py
you can eval our network with this filetest/kinship_test.py
you can test our network with this filemodel
the trained weights of our networkdataset
kinfaceW-I kinfaceW-II tskinface datasetsconv
@article{Huang2022,
author = {Sheng Huang and Jingkai Lin and Yun Xin},
title = {Adaptively Weighted k-tuple Metric Network for Kinship Verification},
booktitle = {IEEE Transactions on Cybernetics},
year = {2022}
}
Sheng Huang, Jingkai Lin, Yun Xing, "Adaptively Weighted k-tuple Metric Network for Kinship Verification". IEEE Transactions on Cybernetics 2022,Accepted. (CCF-B)