Attentive occlusion-adaptive deep network for facial landmark detection
作者：
Highlights：
• We introduced the attention module consisting of Channel-wise Attention (CA) and Spatial Attention (SA) to improve its ability to deal with the occlusion and enhance feature representation ability simultaneously.
• Ablation study proves the importance of each module of our proposed model.
• Our proposed methodology reduces the number of entire network parameters, which effectually decreases training time and cost. So, the proposed model is more suitable for scalable data processing. Experimental results prove the better performance of proposed AODN on challenging benchmark datasets.
摘要
•We introduced the attention module consisting of Channel-wise Attention (CA) and Spatial Attention (SA) to improve its ability to deal with the occlusion and enhance feature representation ability simultaneously.•Ablation study proves the importance of each module of our proposed model.•Our proposed methodology reduces the number of entire network parameters, which effectually decreases training time and cost. So, the proposed model is more suitable for scalable data processing. Experimental results prove the better performance of proposed AODN on challenging benchmark datasets.
论文关键词：Facial landmarks detection,Channel-wise attention,Spatial attention,Deep learning,Face alignment
论文评审过程：Received 16 July 2020, Revised 20 December 2021, Accepted 22 December 2021, Available online 27 December 2021, Version of Record 28 January 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.108510
