The concept of self is a fundamental characteristic of the human mind,and the alteration of self is thought to be a core deficit of schizophrenia.Previous studies have demonstrated that patients with schizophrenia are...The concept of self is a fundamental characteristic of the human mind,and the alteration of self is thought to be a core deficit of schizophrenia.Previous studies have demonstrated that patients with schizophrenia are deficient in self-face recognition.Because self faces are not only self-related but also highly familiar,it is unclear whether such deficit arises from the breakdown of the self-awareness or the failure of recognizing the familiarity of self faces.Here we directly tested these two alternatives by instructing patients with schizophrenia to recognize the identity of a morphed face created by blending face features between any of two identities from the self face,a familiar face,and a novel face.We found that there was no association between the recognition of the self and the recognition of the familiarity,suggesting these two component processes are independent in schizophrenia.Further,patients with schizophrenia were significantly worse in recognizing the familiarity of faces than normal participants,whereas no difference in the sense of self was found between the two groups.Taken together,our finding suggests that it is the sense of familiarity,not the sense of self,that is selectively impaired in self-face recognition in schizophrenia.Thus,our study challenges the hypothesis that the deficit in self-face recognition in schizophrenia reflects the breakdown of self-awareness.展开更多
非约束环境下人脸图像具有背景复杂、尺度分布广泛等特点,当前检测器在标签分配和特征提取方面分别存在人脸匹配锚点数量不均衡和卷积核增长视野受限的问题,导致网络难以进行细粒度优化.针对上述问题,文中提出基于锚点损失优化的细粒度...非约束环境下人脸图像具有背景复杂、尺度分布广泛等特点,当前检测器在标签分配和特征提取方面分别存在人脸匹配锚点数量不均衡和卷积核增长视野受限的问题,导致网络难以进行细粒度优化.针对上述问题,文中提出基于锚点损失优化的细粒度人脸检测方法(Fine-Grained Face Detection Method Based on Anchor Loss Optimization,FALO).首先,分析人脸匹配锚点数量与损失的关系,提出锚点损失优化算法,细粒度地调整训练中分类与定位损失.然后,设计上下文特征融合模块,在背景中有效提取多尺度特征.最后,综合考虑卷积神经网络和自注意力机制,构造自注意力辅助分支,补充检测器感受野的同时提高对不同纵横比人脸的注意力.在多个数据集上的实验表明,FALO可兼顾实时计算效率和高精度检测,在困难样本挖掘中具有一定优势.展开更多
基金supported by the 100 Talents Program of the Chinese Academy of Sciencesthe National Natural Science Foundation of China (91132703)+2 种基金the National Basic Research Program of China(2010CB833903, 2011CB505402)the China Postdoctoral Science Foundation (20060400367)the Fundamental Research Funds for Central Universities
文摘The concept of self is a fundamental characteristic of the human mind,and the alteration of self is thought to be a core deficit of schizophrenia.Previous studies have demonstrated that patients with schizophrenia are deficient in self-face recognition.Because self faces are not only self-related but also highly familiar,it is unclear whether such deficit arises from the breakdown of the self-awareness or the failure of recognizing the familiarity of self faces.Here we directly tested these two alternatives by instructing patients with schizophrenia to recognize the identity of a morphed face created by blending face features between any of two identities from the self face,a familiar face,and a novel face.We found that there was no association between the recognition of the self and the recognition of the familiarity,suggesting these two component processes are independent in schizophrenia.Further,patients with schizophrenia were significantly worse in recognizing the familiarity of faces than normal participants,whereas no difference in the sense of self was found between the two groups.Taken together,our finding suggests that it is the sense of familiarity,not the sense of self,that is selectively impaired in self-face recognition in schizophrenia.Thus,our study challenges the hypothesis that the deficit in self-face recognition in schizophrenia reflects the breakdown of self-awareness.
文摘非约束环境下人脸图像具有背景复杂、尺度分布广泛等特点,当前检测器在标签分配和特征提取方面分别存在人脸匹配锚点数量不均衡和卷积核增长视野受限的问题,导致网络难以进行细粒度优化.针对上述问题,文中提出基于锚点损失优化的细粒度人脸检测方法(Fine-Grained Face Detection Method Based on Anchor Loss Optimization,FALO).首先,分析人脸匹配锚点数量与损失的关系,提出锚点损失优化算法,细粒度地调整训练中分类与定位损失.然后,设计上下文特征融合模块,在背景中有效提取多尺度特征.最后,综合考虑卷积神经网络和自注意力机制,构造自注意力辅助分支,补充检测器感受野的同时提高对不同纵横比人脸的注意力.在多个数据集上的实验表明,FALO可兼顾实时计算效率和高精度检测,在困难样本挖掘中具有一定优势.
