ChildPredictor: A Child Face Prediction Framework with Disentangled Learning - Xuehui Wang (王雪辉)

News

Apr. 15, 2022: We release the trained models with samples for ChildPredictor.
Mar. 31, 2022: The paper is accepted by the IEEE Transactions on Multimedia.
Feb. 8, 2022: We release the code for ChildPredictor. We are considerring to release the original data of the collected FF-Database.

FF-Database

We will release the larger-than-ever kinship dataset (FF-Database) after the publication.

The data collection pipeline is shown as follows:
Some families are shown as follows:

Network Architectures

network

The whole network architecture of ChildPredictor.

network_x_domain

The network architecture of $X$ domain.

network_y_domain

The network architecture of $Y$ domain.

Results on Real Families

The generated results on the collected FF-Database:
The generated results on other datasets:
The disentangled learning analysis is as:

Disentangled learning of $X$ domain.

disentangled_learning_y

Disentangled learning of $Y$ domain.

The ablation study is as:

Please refer to Official Github Page for more implementation details and high-fidelity images.
Lai-Man Po and Wei Shen are the PhD advisor of the first and third author, respectively.

Citation

@Article{Zhao_2022_TMM_ChildPredictor,
        title={ChildPredictor: A Child Face Prediction Framework with Disentangled Learning},
        author={Zhao, Yuzhi and Po, Lai-Man and Wang, Xuehui and Yan, Qiong and Shen, Wei and Zhang, Yujia and Liu, Wei and Wong Chun-Kit and Pang, Chiu-Sing and Ou, Weifeng and Yu, Wing-Yin and Liu, Buhua},
        journal={IEEE Transactions on Multimedia (TMM)},
        year={2022}
}

Contact