基于嵌套残差注意力模块的人物姿态转换方法

Nested Residual Attention Module-Based for Human Pose Transfer

近几年,人们围绕人物图像合成技术展开了多项研究,姿态转换就是其中一个。作为条件输入的姿态信息的引导有局限性,视角变换时生成模型难以处理复杂的人物外观特征。注意力机制可以有效提取图像中的重要部分,通过将提取特征用的残差块嵌入到残差注意力模块中,通过短跳跃连接来逐步学习姿态相关性,自适应地选择空间像素,充分利用姿态转换过程中的全局空间信息,提高生成网络的表征能力,生成具有目标姿态的高质量人物图像。在多类别大型服装数据集DeepFashion上进行测试,验证了所提出算法的有效性。

In recent years, several researches have been carried out around the techniques of human image synthesis, and pose transfer is one of them. The guidance of pose information as conditional input has limitations, and it is difficult to generate models to handle complex character appearance features during perspective transformation. The attention mechanism can effectively extract the important parts of the image, learn the pose correlation step by step by through the short skip connections embedding the residual blocks for feature extraction into the residual attention module, select spatial pixels adaptively making full use of the global spatial information in the pose transformation process, improve the representational capability of the generative network and generate high-quality person images with target pose. The effectiveness of the proposed algorithm is verified by testing on a multi-category large clothing dataset DeepFashion.