Improving Artistic Design With KFDeformNet: Single-Frame Three-Dimensional Cartoon Face Recovery

We propose KFDeformNet, a novel framework for enhancing 3D cartoon face recovery from single-frame inputs. Existing methods struggle to capture the fine details and exaggerated characteristics common in artistic cartoon designs. KFDeformNet addresses this with two key components: the Multi-level Att...

Full description

Saved in:
Bibliographic Details
Main Authors: Tong Sun, Xiaohui Wang, Yichen Qi
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10926190/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We propose KFDeformNet, a novel framework for enhancing 3D cartoon face recovery from single-frame inputs. Existing methods struggle to capture the fine details and exaggerated characteristics common in artistic cartoon designs. KFDeformNet addresses this with two key components: the Multi-level Attention Keypoint Estimator (MAKEstimator) and the Fourier Representation Deformer (FRDeformer). MAKEstimator improves 2D facial keypoint estimation by utilizing multi-level channel-wise aggregation and the Swin Transformer, overcoming challenges in keypoint localization. FRDeformer uses Fourier representation to project 3D template mesh points to high-order space, preserving geometric details and enriching the 3D representation. We evaluate KFDeformNet on various cartoon face datasets, demonstrating that MAKEstimator outperforms state-of-the-art methods in keypoint precision and FRDeformer achieves better 3D recovery with reduced MSE and faster convergence, aided by the Unified Layer Hyper-Network (ULHN) structure. Our results show that KFDeformNet surpasses existing methods in both accuracy and efficiency, enabling improved 3D cartoon face recovery for creative applications.
ISSN:2169-3536