Real-time 3D facial tracking via cascaded compositional learning
Research output: Contribution to journal › Article › peer-review
We further deeply investigate the effect of synthesized facial images on training non-deep learning methods such as GoMBF-Cascade for 3D facial tracking. We apply three types synthetic images with various naturalness levels for training two different tracking methods, and compare the performance of the tracking models trained on real data, on synthetic data and on a mixture of data. The experimental results indicate that, i) the model trained purely on synthetic facial imageries can hardly generalize well to unconstrained real-world data, ii) involving synthetic faces into training benefits tracking in some certain scenarios but degrades the tracking model’s generalization ability. These two insights could benefit a range of non-deep learning facial image analysis tasks where the labelled real data is difficult to acquire.
|Journal||IEEE Transactions on Image Processing|
|Publication status||Accepted for publication - 8 Mar 2021|
- Real-time 3D Facial Tracking via Cascaded Compositional Learning_pp
Rights statement: The embargo end date of 2050 is a temporary measure until we know the publication date. Once we know the publication date the full text of this article will be able to view shortly afterwards.
Accepted author manuscript (Post-print), 1.79 MB, PDF document
Due to publisher’s copyright restrictions, this document is not freely available to download from this website until: 1/01/50