MFFNet: Single facial depth map refinement using multi-level feature fusion

作者：

Highlights：

• A novel multi-level feature fusion CNN for facial depth map refinement.

• Local multi-level feature fusion (LMLF) block and inter-stage skip connections for smoothing the noise and boosting detailed facial structure.

• An effective data augmentation method that synthesizing noisy facial depth maps of various poses.

• Outperform state-of-the-art methods on multiple datasets in terms of quantitative metrics and visual results.

摘要

•A novel multi-level feature fusion CNN for facial depth map refinement.•Local multi-level feature fusion (LMLF) block and inter-stage skip connections for smoothing the noise and boosting detailed facial structure.•An effective data augmentation method that synthesizing noisy facial depth maps of various poses.•Outperform state-of-the-art methods on multiple datasets in terms of quantitative metrics and visual results.

论文关键词：Depth refinement,Facial depth map,Feature fusion

论文评审过程：Received 13 September 2020, Revised 8 January 2022, Accepted 24 January 2022, Available online 31 January 2022, Version of Record 11 February 2022.

论文官网地址：https://doi.org/10.1016/j.image.2022.116649