Mutual information maximizing GAN inversion for real face with identity preservation

Recent generative adversarial networks (GANs) have yielded remarkable performance in face image synthesis. GAN inversion embeds an image into the latent space of a pretrained generator, enabling it to be used for real face manipulation. However, current inversion approaches for real faces suffer the dilemma of initialization collapse and identity loss. In this paper, we propose a hierarchical GAN inversion for real faces with identity preservation based on mutual information maximization. We first use a facial domain guaranteed initialization to avoid the initialization collapse. Furthermore, we prove that maximizing the mutual information between inverted faces and their identities is equivalent to minimizing the distance between identity features from inverted and original faces. Optimization for real face inversion with identity preservation is implemented on this mutual information-maximizing constraint. Extensive experimental results show that our approach outperforms state-of-the-art solutions for inverting and editing real faces, particularly in terms of face identity preservation.     

安卓平台上的医用图像信息采集系统的开发设计

为了探讨在安卓平台上构建医用图像采集系统的开发个案,分析通过以智能手机、平板电脑为核心安卓设备通过拍照获得化验单数据后进行文本识别并提交智慧医疗系统的解决方案。本文首先通过二值化算法形成低阈值图像数据,使用卷积神经元网络算法对文本进行逐一识别,使用K-means算法对识别后的单字文本进行字段记录值的整合并形成元数据库服务于其他智慧医疗系统模块。在使用9000组数据对神经元网络进行前期训练的前提下,该系统的识别准确率达到了99.5%以上。本系统具有一定的可行性,对未来智慧医疗的系统开发有实践意义。     

Emotion recognition at the edge with AI specific low power architectures

Nowadays Deep Learning is applied in almost every research field and helps getting amazing results in a great number of challenging tasks. The main problem is that this kind of learning and consequently Neural Networks that can be defined deep, are resource intensive. They need specialized hardware to perform computation in a reasonable time. Many tasks are mandatory to be as much real-time as possible . It is needed to optimize many components such as code, algorithms, numeric accuracy and hardware, to make them “efficient and usable”. All these optimizations can help us to produce incredibly accurate and fast learning models. The paper reports a study in this direction for the challenging face detection and emotion recognition tasks.     

基于深度概率图模型的鲁棒人脸画像合成

针对基于数据驱动的人脸画像合成算法像素特征缺乏对光照变化和复杂背景的鲁棒性,常合成低质量的画像的问题,文中提出基于深度概率图模型的鲁棒人脸画像合成算法.采用预处理方法调整测试照片的光照亮度和人脸姿态,使之与训练照片一致.采用深度特征代替像素特征进行近邻匹配,采用深度概率图模型对画像重建权重和深度特征权重联合建模,得到合成画像的最佳重构表示.为了提高画像合成速度,提出快速近邻搜索方法.实验验证文中算法的鲁棒性和快速性.     

扫描地图的计算机自动识别

扫描地图的计算机自动识别是当今地图制图学科的前沿课题，也是其重要的研究方向。为此，本文首先将模式识别技术引入到地图制图领域中，提出了四种地图符号的计算机描述方法，然后又讨论了扫描地图的信息获取与处理，并具体地分析了扫描地图的计算机识别方法及其识别系统．最后还给出了相应的符号识别实验。     

Influence of ozone-limonene reactions on perceived air quality

This study conducted short-term assessments of perceived air quality (PAQ) for six different realistic concentrations of ozone and limonene, separately or together, in room air. The impact of filtration and the influence of the ozone generation method were also examined. The evaluations were made in four identical 40 m3 low-polluting test offices ventilated at 1.4 h(-1) or in two identical 30 m3 stainless-steel chambers ventilated at 1.9 h(-1). Concentrations of ozone, total volatile organic compounds and size-fractionated particles were continuously monitored in each experiment. The results indicate that, for each of the six conditions, the PAQ was poorer when ozone and limonene were present together compared with when only ozone or only limonene was present. In the test offices a correlation was observed between the number of secondary organic aerosols produced by a given ozone/limonene condition and the sensory pollution load for that condition. The particles themselves do not appear to be the primary causative agents, but instead are co-varying surrogates for sensory offending gas-phase species. PRACTICAL IMPLICATIONS: Although the health consequences of long-term exposures to the products of ozone-initiated indoor chemistry remain to be determined, we judge that the sensory offending nature of selected products provides an additional reason to limit indoor ozone levels. Devices that emit ozone at significant rates should not be used indoors. Ozone-filtration of make-up air should also be beneficial in mechanically ventilated buildings located in regions that repeatedly violate outdoor ozone standards. Additionally, the use of limonene containing products should be curtailed during periods when indoor ozone levels are elevated.     

汽车牌照自动识别系统在公路收费口的应用

最近及未来的几年 ,上海市境内将建成多条高速公路 ,如何提高高速公路的服务质量 ,出入口的通行能力是一条比较重要的衡量标准。汽车牌照自动识别系统利用先进的摄像技术、图像识别技术和计算机技术 ,对抓拍图像进行正确分割和快速处理 ,自动识别出汽车的牌照号码 ,提升了高速公路的服务水平     

注册建筑师制度在中国的建立及管理统计

文章重点从政策层面比较系统地回顾了中国建设行业注册建筑师执业管理制度从建立到发展演变的过程以及注册资格国际互认方面的情况。     

Effect of seepage force on tunnel face stability reinforced with multi-step pipe grouting

Tunneling in difficult geological conditions is often inevitable especially in urban areas. Ground improvement and reinforcement techniques are required to guarantee safe tunnel excavations and/or to prevent damage to adjacent structures. The steel pipe-reinforced multi-step grouting method has been recently applied to tunnel sites as an auxiliary technique in Korea for impermeabilization in underwater tunnels as well as for reinforcement. However, this technique has been usually employed empirically without much understanding with regard to its effect on the tunnel safety. In this study, the face stability with steel pipe-reinforced multi-step grouting in underwater tunnels was evaluated by simultaneously considering two factors: one is the effective stress acting on the tunnel face calculated by limit theorem and limit equilibrium method; the other is the seepage force obtained by means of numerical analysis. This study revealed that the influence of the steel pipe-reinforced multi-step grouting on the support pressure required for the stability of the tunnel face in dry condition is not significant while there is relatively a significant reduction in seepage forces by adopting the technique in the underwater tunnel. The effect of permeability anisotropy on the seepage force acting on the tunnel face was also assessed by conducting a coupled analysis.     

Vector projection for face recognition

In this paper, a novel approach for face recognition is proposed by using vector projection length to formulate the pattern recognition problem. Face images of a single-object class are more similar than those of different-object classes. The projection length of a test image vector on the direction of a training image vector can measure the similarity of the two images. But the decision cannot be made by only a training image which is the most similar to the test one, the mean image vector of each class also contributes to the final classification. Thus, the decision of the proposed vector projection classification (VPC) algorithm is ruled in favor of the maximum combination projection length. To address the partial occlusion problem in face recognition, we propose a local vector projection classification (LVPC) algorithm. The experimental results show that the proposed VPC and LVPC approaches are efficient and outperform some existing approaches.