Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00067
I. Masi, Yuehua Wu, Tal Hassner, P. Natarajan
Face recognition made tremendous leaps in the last five years with a myriad of systems proposing novel techniques substantially backed by deep convolutional neural networks (DCNN). Although face recognition performance sky-rocketed using deep-learning in classic datasets like LFW, leading to the belief that this technique reached human performance, it still remains an open problem in unconstrained environments as demonstrated by the newly released IJB datasets. This survey aims to summarize the main advances in deep face recognition and, more in general, in learning face representations for verification and identification. The survey provides a clear, structured presentation of the principal, state-of-the-art (SOTA) face recognition techniques appearing within the past five years in top computer vision venues. The survey is broken down into multiple parts that follow a standard face recognition pipeline: (a) how SOTA systems are trained and which public data sets have they used; (b) face preprocessing part (detection, alignment, etc.); (c) architecture and loss functions used for transfer learning (d) face recognition for verification and identification. The survey concludes with an overview of the SOTA results at a glance along with some open issues currently overlooked by the community.
{"title":"Deep Face Recognition: A Survey","authors":"I. Masi, Yuehua Wu, Tal Hassner, P. Natarajan","doi":"10.1109/SIBGRAPI.2018.00067","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00067","url":null,"abstract":"Face recognition made tremendous leaps in the last five years with a myriad of systems proposing novel techniques substantially backed by deep convolutional neural networks (DCNN). Although face recognition performance sky-rocketed using deep-learning in classic datasets like LFW, leading to the belief that this technique reached human performance, it still remains an open problem in unconstrained environments as demonstrated by the newly released IJB datasets. This survey aims to summarize the main advances in deep face recognition and, more in general, in learning face representations for verification and identification. The survey provides a clear, structured presentation of the principal, state-of-the-art (SOTA) face recognition techniques appearing within the past five years in top computer vision venues. The survey is broken down into multiple parts that follow a standard face recognition pipeline: (a) how SOTA systems are trained and which public data sets have they used; (b) face preprocessing part (detection, alignment, etc.); (c) architecture and loss functions used for transfer learning (d) face recognition for verification and identification. The survey concludes with an overview of the SOTA results at a glance along with some open issues currently overlooked by the community.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129247177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00034
Ron Vanderfeesten, J. Bikker
We present an algorithm for generating procedural displacement maps for wrinkle patterns measured from photographs or scans. These displacement maps can contain wrinkle patterns that appear at the meso-and microscale, and are modeled using several spatially varying parameters such as the size, shape and distribution of each individual skin wrinkle. We present an algorithm to measure the parameters of skin wrinkle patterns, and show how to adapt the measured parameters to generate displacement maps with similar properties for 3D models other than the one measured. Lastly, we evaluate the quality of the generated maps by comparing them to maps acquired by scanning human skin.
{"title":"Example-Based Skin Wrinkle Displacement Maps","authors":"Ron Vanderfeesten, J. Bikker","doi":"10.1109/SIBGRAPI.2018.00034","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00034","url":null,"abstract":"We present an algorithm for generating procedural displacement maps for wrinkle patterns measured from photographs or scans. These displacement maps can contain wrinkle patterns that appear at the meso-and microscale, and are modeled using several spatially varying parameters such as the size, shape and distribution of each individual skin wrinkle. We present an algorithm to measure the parameters of skin wrinkle patterns, and show how to adapt the measured parameters to generate displacement maps with similar properties for 3D models other than the one measured. Lastly, we evaluate the quality of the generated maps by comparing them to maps acquired by scanning human skin.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"2014 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128060118","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00024
D. Coimbra, T. T. A. T. Neves, A. Telea, F. Paulovich
The development of multimedia and network technologies strongly increase the interest on Internet broadcasting or streaming services, especially for soccer games. An example is the 2014 World Cup soccer tournament that registered record-breaking audiences worldwide, providing attractive alternatives to traditional TV viewing. The confluence of video streaming and computational resources opens up many possibilities for applications such as the online detection of interesting events, strategy analysis, or statistics comparisons. While much research targets algorithms to detect match statistics, strategy, retrieval, and indexing, the problem of presenting such information to users is much less studied. This paper proposes a simple but effective visual metaphor to help users browse and get insight into sports matches, with a focus on soccer games. We extract video segments, based on audio and metadata, identifying the main events of a game. Next, we use such events to define a visual representation that preserves the time-order of the video sequence, highlighting the most important events. Our visual representation enables the quick finding of the main events, allowing users to improve navigation when exploring a match, and also provides a way to evaluate the quality of a game or entire tournaments. We demonstrate our approach by applying it to several matches of 2014 World Cup, analyzing its knockout stage and comparing the final match in six different languages.
{"title":"The Shape of the Game","authors":"D. Coimbra, T. T. A. T. Neves, A. Telea, F. Paulovich","doi":"10.1109/SIBGRAPI.2018.00024","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00024","url":null,"abstract":"The development of multimedia and network technologies strongly increase the interest on Internet broadcasting or streaming services, especially for soccer games. An example is the 2014 World Cup soccer tournament that registered record-breaking audiences worldwide, providing attractive alternatives to traditional TV viewing. The confluence of video streaming and computational resources opens up many possibilities for applications such as the online detection of interesting events, strategy analysis, or statistics comparisons. While much research targets algorithms to detect match statistics, strategy, retrieval, and indexing, the problem of presenting such information to users is much less studied. This paper proposes a simple but effective visual metaphor to help users browse and get insight into sports matches, with a focus on soccer games. We extract video segments, based on audio and metadata, identifying the main events of a game. Next, we use such events to define a visual representation that preserves the time-order of the video sequence, highlighting the most important events. Our visual representation enables the quick finding of the main events, allowing users to improve navigation when exploring a match, and also provides a way to evaluate the quality of a game or entire tournaments. We demonstrate our approach by applying it to several matches of 2014 World Cup, analyzing its knockout stage and comparing the final match in six different languages.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128929255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00065
João Macedo, F. Costa, J. A. D. Santos
The acquisition and distribution of child sexual content are some of the most important concerns for legislative systems and law enforcement agencies around the world. There is a great demand for automatic detection of child pornography, mainly due to the large amount of existent data and the facility someone can share this content over the internet. Although there are some proposed methods to automatically detect child pornography content in the literature, there is no available dataset to assess and compare the performance of these methods due to legal restrictions, considering that in many countries the distribution or possession of this material is a crime by Law. To mitigate this problem, we work with the Brazilian Federal Police to structure and organize a benchmark methodology for child pornography to make it possible the comparison of distinct categories of child pornography detectors. Therefore, we present in this paper the used methodology for the creation of a new annotated dataset of images of child pornography. We also propose a child pornography detection step-wise methodology based on automatic age estimation combined with a pornography detector, which is evaluated using the described benchmark dataset. The proposed approach achieved results (79.84% accuracy) that overcome two tools currently used by the Brazilian Federal Police.
{"title":"A Benchmark Methodology for Child Pornography Detection","authors":"João Macedo, F. Costa, J. A. D. Santos","doi":"10.1109/SIBGRAPI.2018.00065","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00065","url":null,"abstract":"The acquisition and distribution of child sexual content are some of the most important concerns for legislative systems and law enforcement agencies around the world. There is a great demand for automatic detection of child pornography, mainly due to the large amount of existent data and the facility someone can share this content over the internet. Although there are some proposed methods to automatically detect child pornography content in the literature, there is no available dataset to assess and compare the performance of these methods due to legal restrictions, considering that in many countries the distribution or possession of this material is a crime by Law. To mitigate this problem, we work with the Brazilian Federal Police to structure and organize a benchmark methodology for child pornography to make it possible the comparison of distinct categories of child pornography detectors. Therefore, we present in this paper the used methodology for the creation of a new annotated dataset of images of child pornography. We also propose a child pornography detection step-wise methodology based on automatic age estimation combined with a pornography detector, which is evaluated using the described benchmark dataset. The proposed approach achieved results (79.84% accuracy) that overcome two tools currently used by the Brazilian Federal Police.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121860982","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00010
L. C. Ribeiro, J. Papa
Dialogue Act classification is a relevant problem for the Natural Language Processing field either as a standalone task or when used as input for downstream applications. Despite its importance, most of the existing approaches rely on supervised techniques, which depend on annotated samples, making it difficult to take advantage of the increasing amount of data available in different domains. In this paper, we briefly review the most commonly used datasets to evaluate Dialogue Act classification approaches and introduce the Optimum-Path Forest (OPF) classifier to this task. Instead of using its original strategy to determine the corresponding class for each cluster, we use a modified version based on majority voting, named M-OPF, which yields good results when compared to k-means and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN), according to accuracy and V-measure. We also show that M-OPF, and consequently OPF, are less sensitive to hyper-parameter tuning when compared to HDBSCAN.
{"title":"Unsupervised Dialogue Act Classification with Optimum-Path Forest","authors":"L. C. Ribeiro, J. Papa","doi":"10.1109/SIBGRAPI.2018.00010","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00010","url":null,"abstract":"Dialogue Act classification is a relevant problem for the Natural Language Processing field either as a standalone task or when used as input for downstream applications. Despite its importance, most of the existing approaches rely on supervised techniques, which depend on annotated samples, making it difficult to take advantage of the increasing amount of data available in different domains. In this paper, we briefly review the most commonly used datasets to evaluate Dialogue Act classification approaches and introduce the Optimum-Path Forest (OPF) classifier to this task. Instead of using its original strategy to determine the corresponding class for each cluster, we use a modified version based on majority voting, named M-OPF, which yields good results when compared to k-means and Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN), according to accuracy and V-measure. We also show that M-OPF, and consequently OPF, are less sensitive to hyper-parameter tuning when compared to HDBSCAN.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"118 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121385241","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00036
Luiza A. Hagemann, M. M. O. Neto
Physically-based rendering systems use proprietary scene description formats. Thus, by selecting a given renderer for the development of a new technique, one is often constrained to test and demonstrate it on the limited set of test scenes available for that particular renderer. This makes it difficult to compare techniques implemented on different systems. We present a solution for automatic conversion among scene description formats used by physically-based rendering systems. It enables algorithms implemented on different renderers to be tested on the same scene, providing better means of assessing their strengths and limitations. Our system can be integrated with development and benchmarking APIs, lending to full orthogonality among algorithms, rendering systems, and scene files.
{"title":"Scene Conversion for Physically-Based Renderers","authors":"Luiza A. Hagemann, M. M. O. Neto","doi":"10.1109/SIBGRAPI.2018.00036","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00036","url":null,"abstract":"Physically-based rendering systems use proprietary scene description formats. Thus, by selecting a given renderer for the development of a new technique, one is often constrained to test and demonstrate it on the limited set of test scenes available for that particular renderer. This makes it difficult to compare techniques implemented on different systems. We present a solution for automatic conversion among scene description formats used by physically-based rendering systems. It enables algorithms implemented on different renderers to be tested on the same scene, providing better means of assessing their strengths and limitations. Our system can be integrated with development and benchmarking APIs, lending to full orthogonality among algorithms, rendering systems, and scene files.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114622309","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00007
Jan Hurtado, M. Gattass, A. Raposo, Jéferson Coêlho
The generation of triangular meshes typically introduces undesired noise which comes from different sources. Mesh denoising is a geometry processing task to remove this kind of distortion. To preserve the geometric fidelity of the desired mesh, a mesh denoising algorithm must maintain the object details while removing artificial high-frequencies from the surface. In this work, we propose a two-step algorithm which uses adaptive patches and bilateral filtering to denoise the normal vector field, and then update vertex positions fitting the faces to the denoised normals. The computation of the adaptive patches is our main contribution. We formulate this computation as local quadratic optimization problems that can be controlled by a set of parameters to obtain the desired behavior. We compared our proposal with several algorithms proposed in the literature using synthetic and real data. Our algorithm yields better results in general and is based on a formal mathematical formulation.
{"title":"Adaptive Patches for Mesh Denoising","authors":"Jan Hurtado, M. Gattass, A. Raposo, Jéferson Coêlho","doi":"10.1109/SIBGRAPI.2018.00007","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00007","url":null,"abstract":"The generation of triangular meshes typically introduces undesired noise which comes from different sources. Mesh denoising is a geometry processing task to remove this kind of distortion. To preserve the geometric fidelity of the desired mesh, a mesh denoising algorithm must maintain the object details while removing artificial high-frequencies from the surface. In this work, we propose a two-step algorithm which uses adaptive patches and bilateral filtering to denoise the normal vector field, and then update vertex positions fitting the faces to the denoised normals. The computation of the adaptive patches is our main contribution. We formulate this computation as local quadratic optimization problems that can be controlled by a set of parameters to obtain the desired behavior. We compared our proposal with several algorithms proposed in the literature using synthetic and real data. Our algorithm yields better results in general and is based on a formal mathematical formulation.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127490172","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2018-10-01DOI: 10.1109/SIBGRAPI.2018.00017
L. T. Goncalves, J. O. Gaya, Paulo Jorge Lilles Drews Junior, S. Botelho
Poor visibility is a common problem when capturing images in participating mediums such as mist or water. The problem of generating a haze-free image based on a hazy one can be described as image dehazing. Previous approaches dealt with this problem using physical models based on priors and simplifications. In this paper, we demonstrate that an end-to-end convolutional neural network is able to learn the dehazing process with no parameters or priors required, resulting in a more generic method. Even though our model is trained entirely with hazy indoor images, we are able to fully restore outdoor images with real haze. Also, we propose an architecture containing the novel Guided Layers, introduced in order to reduce the loss of spatial information while restoring the images. Our method outperforms other machine learning based models, yielding superior results both qualitatively and quantitatively.
{"title":"GuidedNet: Single Image Dehazing Using an End-to-End Convolutional Neural Network","authors":"L. T. Goncalves, J. O. Gaya, Paulo Jorge Lilles Drews Junior, S. Botelho","doi":"10.1109/SIBGRAPI.2018.00017","DOIUrl":"https://doi.org/10.1109/SIBGRAPI.2018.00017","url":null,"abstract":"Poor visibility is a common problem when capturing images in participating mediums such as mist or water. The problem of generating a haze-free image based on a hazy one can be described as image dehazing. Previous approaches dealt with this problem using physical models based on priors and simplifications. In this paper, we demonstrate that an end-to-end convolutional neural network is able to learn the dehazing process with no parameters or priors required, resulting in a more generic method. Even though our model is trained entirely with hazy indoor images, we are able to fully restore outdoor images with real haze. Also, we propose an architecture containing the novel Guided Layers, introduced in order to reduce the loss of spatial information while restoring the images. Our method outperforms other machine learning based models, yielding superior results both qualitatively and quantitatively.","PeriodicalId":208985,"journal":{"name":"2018 31st SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126411420","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：