结合多尺度融合特征和残差注意力机制的 联合三维人脸重建及密集对齐算法.

Aiming at the problem of insufficient accuracy of 3D face reconstruction and dense alignment algorithm,this paper introduced densely connected multi-scale feature fusion module and residual attention mechanism to design a powerful network. For the encoder structure, this paper introduced a densely connected multi-scale feature fusion module to obtain multi-scale fusion features, which made the encoder obtain richer information. The decoder module introduced residual attention mechanism to strengthen the network 's attention to important features while suppressing unnecessary noise. According to the experimental results, compared with other algorithms, the proposed algorithm achieves a significant improvement. Compared with the PRN et, the algorithm uses fewer parameters to achieve a performance improvement of 7. 7% to 12. 1 % in various indicators. [ABSTRACT FROM AUTHOR]