The needs for photo-realistic modelling of the complete details, and geometrically accurate 3D models are growing rapidly in several fields, especially in engineering and cultural heritage documentation. This paper ex...The needs for photo-realistic modelling of the complete details, and geometrically accurate 3D models are growing rapidly in several fields, especially in engineering and cultural heritage documentation. This paper explored the geometry of the Pictometry images （vertical and oblique） and the possibility of using this imagery in 3D modelling to produce photo-realistic and accurate models. In addition, merging terrestrial imagery with Pictometry imagery to get more ground level details has been investigated. All work has been carried out using the available software packages at the IESSG （Institute of Engineering Surveying and Space Geodesy） and using data provided by Blom Aerofilms Ltd. The results of the aerial triangulation of different Pictometry blocks showed that high quality image measurements have been achieved for all the image blocks. Extraction of 3D geometry for all buildings in the study area has been performed using both vertical Pictometry imagery and UltraCamD imagery. The successful combining of vertical and oblique Pictometry images provided an excellent opportunity to produce an efficient method of high quality urban model texturing. The integration of terrestrial images of building facades （whose texture needs enhancement） with the combined aerial imagery block has been successfully and automatically performed.展开更多
The photo-reactions between metabolic products of human sweat and dyestuffs on garments may produce many toxic substances which could directly contact skin and threaten human health. In order to investigate the impact...The photo-reactions between metabolic products of human sweat and dyestuffs on garments may produce many toxic substances which could directly contact skin and threaten human health. In order to investigate the impact of the perspiration on photo-fading of reactive dyes on cellulose, nine commercial reactive dyes belonging to three types of chromophores (azo, Cu-complex azo and anthraquinone) respectively were chosen and their perspiration-light stability on cotton fabric was studied following ATTS test standard. It is found that the impact of the artificial perspiration on dyes varies with different chromophores: anthraquinone reactive dyes always show the best photo-stability, whereas Cu-complex azo reactive dyes appear to be the most sensitive under exposure to light and perspiration. The pH value of perspiration also greatly influences the fading of dyes with different reactive groups: the fading rate of most chlorotriazinyl reactive dyes in acidic perspiration (pH=3.5) is higher than in alkaline perspiration (pH=8.0), while the reverse is true for most vinylsulphonyl dyes. Furthermore, the study of the contribution of individual component of the artificial perspiration discloses that L-Histidine monohydrochloride monohydrate, DL-Aspartic acid and lactic acid play the major roles on the photo-fading of those selected dyestuffs and inorganic salts including disodium hydrogen phosphate and sodium chloride usually decelerate photo-fading.展开更多
The technique of facial attribute manipulation has found increasing application,but it remains challenging to restrict editing of attributes so that a face’s unique details are preserved.In this paper,we introduce ou...The technique of facial attribute manipulation has found increasing application,but it remains challenging to restrict editing of attributes so that a face’s unique details are preserved.In this paper,we introduce our method,which we call a mask-adversarial autoencoder(M-AAE).It combines a variational autoencoder(VAE)and a generative adversarial network(GAN)for photorealistic image generation.We use partial dilated layers to modify a few pixels in the feature maps of an encoder,changing the attribute strength continuously without hindering global information.Our training objectives for the VAE and GAN are reinforced by supervision of face recognition loss and cycle consistency loss,to faithfully preserve facial details.Moreover,we generate facial masks to enforce background consistency,which allows our training to focus on the foreground face rather than the background.Experimental results demonstrate that our method can generate high-quality images with varying attributes,and outperforms existing methods in detail preservation.展开更多
文摘The needs for photo-realistic modelling of the complete details, and geometrically accurate 3D models are growing rapidly in several fields, especially in engineering and cultural heritage documentation. This paper explored the geometry of the Pictometry images （vertical and oblique） and the possibility of using this imagery in 3D modelling to produce photo-realistic and accurate models. In addition, merging terrestrial imagery with Pictometry imagery to get more ground level details has been investigated. All work has been carried out using the available software packages at the IESSG （Institute of Engineering Surveying and Space Geodesy） and using data provided by Blom Aerofilms Ltd. The results of the aerial triangulation of different Pictometry blocks showed that high quality image measurements have been achieved for all the image blocks. Extraction of 3D geometry for all buildings in the study area has been performed using both vertical Pictometry imagery and UltraCamD imagery. The successful combining of vertical and oblique Pictometry images provided an excellent opportunity to produce an efficient method of high quality urban model texturing. The integration of terrestrial images of building facades （whose texture needs enhancement） with the combined aerial imagery block has been successfully and automatically performed.
基金Supported by Foundation of University Doctoral Programby the State Ministry of Education (No.20050255002)
文摘The photo-reactions between metabolic products of human sweat and dyestuffs on garments may produce many toxic substances which could directly contact skin and threaten human health. In order to investigate the impact of the perspiration on photo-fading of reactive dyes on cellulose, nine commercial reactive dyes belonging to three types of chromophores (azo, Cu-complex azo and anthraquinone) respectively were chosen and their perspiration-light stability on cotton fabric was studied following ATTS test standard. It is found that the impact of the artificial perspiration on dyes varies with different chromophores: anthraquinone reactive dyes always show the best photo-stability, whereas Cu-complex azo reactive dyes appear to be the most sensitive under exposure to light and perspiration. The pH value of perspiration also greatly influences the fading of dyes with different reactive groups: the fading rate of most chlorotriazinyl reactive dyes in acidic perspiration (pH=3.5) is higher than in alkaline perspiration (pH=8.0), while the reverse is true for most vinylsulphonyl dyes. Furthermore, the study of the contribution of individual component of the artificial perspiration discloses that L-Histidine monohydrochloride monohydrate, DL-Aspartic acid and lactic acid play the major roles on the photo-fading of those selected dyestuffs and inorganic salts including disodium hydrogen phosphate and sodium chloride usually decelerate photo-fading.
基金partially funded by the National Natural Science Foundation of China(No.61972157)the National Social Science Foundation of China(No.18ZD22)+3 种基金the Science and Technology Commission of Shanghai Municipality Program(No.18D1205903)the Science and Technology Commission of Pudong Municipality Program(No.PKJ2018-Y46)the Multidisciplinary Project of Shanghai Jiao Tong University(No.ZH2018ZDA25)partially supported by a joint project of SenseTime and Shanghai Jiao Tong University。
文摘The technique of facial attribute manipulation has found increasing application,but it remains challenging to restrict editing of attributes so that a face’s unique details are preserved.In this paper,we introduce our method,which we call a mask-adversarial autoencoder(M-AAE).It combines a variational autoencoder(VAE)and a generative adversarial network(GAN)for photorealistic image generation.We use partial dilated layers to modify a few pixels in the feature maps of an encoder,changing the attribute strength continuously without hindering global information.Our training objectives for the VAE and GAN are reinforced by supervision of face recognition loss and cycle consistency loss,to faithfully preserve facial details.Moreover,we generate facial masks to enforce background consistency,which allows our training to focus on the foreground face rather than the background.Experimental results demonstrate that our method can generate high-quality images with varying attributes,and outperforms existing methods in detail preservation.
