Yifang Li, Nishant Vishwamitra, Bart P. Knijnenburg, Hongxin Hu, Kelly E. Caine
Computer vision can lead to privacy issues such as unauthorized disclosure of private information and identity theft, but it may also be used to preserve user privacy. For example, using computer vision, we may be able to identify sensitive elements of an image and obfuscate those elements thereby protecting private information or identity. However, there is a lack of research investigating the effectiveness of applying obfuscation techniques to parts of images as a privacy enhancing technology. In particular, we know very little about how well obfuscation works for human viewers or users' attitudes towards using these mechanisms. In this paper, we report results from an online experiment with 53 participants that investigates the effectiveness two exemplar obfuscation techniques: "blurring" and "blocking", and explores users' perceptions of these obfuscations in terms of image satisfaction, information sufficiency, enjoyment, and social presence. Results show that although "blocking" is more effective at de-identification compared to "blurring" or leaving the image "as is", users' attitudes towards "blocking" are the most negative, which creates a conflict between privacy protection and users' experience. Future work should explore alternative obfuscation techniques that could protect users' privacy and also provide a good viewing experience.
{"title":"Blur vs. Block: Investigating the Effectiveness of Privacy-Enhancing Obfuscation for Images","authors":"Yifang Li, Nishant Vishwamitra, Bart P. Knijnenburg, Hongxin Hu, Kelly E. Caine","doi":"10.1109/CVPRW.2017.176","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.176","url":null,"abstract":"Computer vision can lead to privacy issues such as unauthorized disclosure of private information and identity theft, but it may also be used to preserve user privacy. For example, using computer vision, we may be able to identify sensitive elements of an image and obfuscate those elements thereby protecting private information or identity. However, there is a lack of research investigating the effectiveness of applying obfuscation techniques to parts of images as a privacy enhancing technology. In particular, we know very little about how well obfuscation works for human viewers or users' attitudes towards using these mechanisms. In this paper, we report results from an online experiment with 53 participants that investigates the effectiveness two exemplar obfuscation techniques: \"blurring\" and \"blocking\", and explores users' perceptions of these obfuscations in terms of image satisfaction, information sufficiency, enjoyment, and social presence. Results show that although \"blocking\" is more effective at de-identification compared to \"blurring\" or leaving the image \"as is\", users' attitudes towards \"blocking\" are the most negative, which creates a conflict between privacy protection and users' experience. Future work should explore alternative obfuscation techniques that could protect users' privacy and also provide a good viewing experience.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"65 1","pages":"1343-1351"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85714754","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}
M. Rahnemoonfar, M. Yari, Abdullah F. Rahman, Richard J. Kline
There is a vital need to map seagrass ecosystems in order to determine worldwide abundance and distribution. Currently there is no established method for mapping the pothole or scars in seagrass. Detection of seagrass with optical remote sensing is challenged by the fact that light is attenuated as it passes through the water column and reflects back from the benthos. Optical remote sensing of seagrass is only possible if the water is shallow and relatively clear. In reality, coastal waters are commonly turbid, and seagrasses can grow under 10 meters of water or even deeper. One of the most precise sensors to map the seagrass disturbance is side scan sonar. Underwater acoustics mapping produces a high definition, two-dimensional sonar image of seagrass ecosystems. This paper proposes a methodology which detects seagrass potholes in sonar images. Side scan sonar images usually contain speckle noise and uneven illumination across the image. Moreover, disturbance presents complex patterns where most segmentation techniques will fail. In this paper, the quality of image is improved in the first stage using adaptive thresholding and wavelet denoising techniques. In the next step, a novel level set technique is applied to identify the pothole patterns. Our method is robust to noise and uneven illumination. Moreover it can detect the complex pothole patterns. We tested our proposed approach on a collection of underwater sonar images taken from Laguna Madre in Texas. Experimental results in comparison with the ground-truth show the efficiency of the proposed method.
{"title":"The First Automatic Method for Mapping the Pothole in Seagrass","authors":"M. Rahnemoonfar, M. Yari, Abdullah F. Rahman, Richard J. Kline","doi":"10.1109/CVPRW.2017.39","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.39","url":null,"abstract":"There is a vital need to map seagrass ecosystems in order to determine worldwide abundance and distribution. Currently there is no established method for mapping the pothole or scars in seagrass. Detection of seagrass with optical remote sensing is challenged by the fact that light is attenuated as it passes through the water column and reflects back from the benthos. Optical remote sensing of seagrass is only possible if the water is shallow and relatively clear. In reality, coastal waters are commonly turbid, and seagrasses can grow under 10 meters of water or even deeper. One of the most precise sensors to map the seagrass disturbance is side scan sonar. Underwater acoustics mapping produces a high definition, two-dimensional sonar image of seagrass ecosystems. This paper proposes a methodology which detects seagrass potholes in sonar images. Side scan sonar images usually contain speckle noise and uneven illumination across the image. Moreover, disturbance presents complex patterns where most segmentation techniques will fail. In this paper, the quality of image is improved in the first stage using adaptive thresholding and wavelet denoising techniques. In the next step, a novel level set technique is applied to identify the pothole patterns. Our method is robust to noise and uneven illumination. Moreover it can detect the complex pothole patterns. We tested our proposed approach on a collection of underwater sonar images taken from Laguna Madre in Texas. Experimental results in comparison with the ground-truth show the efficiency of the proposed method.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"16 1","pages":"267-274"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85996780","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}
This paper aims to introduce a new vehicle type classification scheme on the images from multi-view surveillance camera. We propose four concepts to increase the performance on the images which have various resolutions from multi-view point. The Deep Learning method is essential to multi-view point image, bagging method makes system robust, data augmentation help to grow the classification capability, and post-processing compensate for imbalanced data. We combine these schemes and build a novel vehicle type classification system. Our system shows 97.84% classification accuracy on the 103,833 images in classification challenge dataset.
{"title":"Vehicle Type Classification Using Bagging and Convolutional Neural Network on Multi View Surveillance Image","authors":"Pyong-Kun Kim, Kil-Taek Lim","doi":"10.1109/CVPRW.2017.126","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.126","url":null,"abstract":"This paper aims to introduce a new vehicle type classification scheme on the images from multi-view surveillance camera. We propose four concepts to increase the performance on the images which have various resolutions from multi-view point. The Deep Learning method is essential to multi-view point image, bagging method makes system robust, data augmentation help to grow the classification capability, and post-processing compensate for imbalanced data. We combine these schemes and build a novel vehicle type classification system. Our system shows 97.84% classification accuracy on the 103,833 images in classification challenge dataset.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"38 1","pages":"914-919"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73096183","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}
The last years have seen a quick rise of digital photography. Plenoptic cameras provide extended capabilities with respect to previous models. Multi-focus cameras enlarge the depth-of-field of the pictures using different focal lengths in the lens composing the array, but questions still arise on how to select and use these lenses. In this work a further insight on the lens selection was made, and a novel method was developed in order to choose the best available lens combination for the disparity estimation. We test different lens combinations, ranking them based on the error and the number of different lenses used, creating a mapping function that relates the virtual depth with the combination that achieves the best result. The results are then organized in a look up table that can be tuned to trade off between performances and accuracy. This allows for fast and accurate lens selection. Moreover, new synthetic images with respective ground truth are provided, in order to confirm that this work performs better than the current state of the art in efficiency and accuracy of the results.
{"title":"Optimizing the Lens Selection Process for Multi-focus Plenoptic Cameras and Numerical Evaluation","authors":"L. Palmieri, R. Koch","doi":"10.1109/CVPRW.2017.223","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.223","url":null,"abstract":"The last years have seen a quick rise of digital photography. Plenoptic cameras provide extended capabilities with respect to previous models. Multi-focus cameras enlarge the depth-of-field of the pictures using different focal lengths in the lens composing the array, but questions still arise on how to select and use these lenses. In this work a further insight on the lens selection was made, and a novel method was developed in order to choose the best available lens combination for the disparity estimation. We test different lens combinations, ranking them based on the error and the number of different lenses used, creating a mapping function that relates the virtual depth with the combination that achieves the best result. The results are then organized in a look up table that can be tuned to trade off between performances and accuracy. This allows for fast and accurate lens selection. Moreover, new synthetic images with respective ground truth are provided, in order to confirm that this work performs better than the current state of the art in efficiency and accuracy of the results.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"33 1","pages":"1763-1774"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73695914","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}
Deep learning based approaches proved to be dramatically effective to address many computer vision applications, including "face recognition in the wild". It has been extensively demonstrated that methods exploiting Deep Convolutional Neural Networks (DCNN) are powerful enough to overcome to a great extent many problems that negatively affected computer vision algorithms based on hand-crafted features. These problems include variations in illumination, pose, expression and occlusion, to mention some. The DCNNs excellent discriminative power comes from the fact that they learn low-and high-level representations directly from the raw image data. Considering this, it can be assumed that the performance of a DCNN are influenced by the characteristics of the raw image data that are fed to the network. In this work, we evaluate the effect of different bounding box dimensions, alignment, positioning and data source on face recognition using DCNNs, and present a thorough evaluation on two well known, public DCNN architectures.
{"title":"Investigating Nuisance Factors in Face Recognition with DCNN Representation","authors":"C. Ferrari, G. Lisanti, S. Berretti, A. Bimbo","doi":"10.1109/CVPRW.2017.86","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.86","url":null,"abstract":"Deep learning based approaches proved to be dramatically effective to address many computer vision applications, including \"face recognition in the wild\". It has been extensively demonstrated that methods exploiting Deep Convolutional Neural Networks (DCNN) are powerful enough to overcome to a great extent many problems that negatively affected computer vision algorithms based on hand-crafted features. These problems include variations in illumination, pose, expression and occlusion, to mention some. The DCNNs excellent discriminative power comes from the fact that they learn low-and high-level representations directly from the raw image data. Considering this, it can be assumed that the performance of a DCNN are influenced by the characteristics of the raw image data that are fed to the network. In this work, we evaluate the effect of different bounding box dimensions, alignment, positioning and data source on face recognition using DCNNs, and present a thorough evaluation on two well known, public DCNN architectures.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"19 1","pages":"583-591"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81111976","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}
Face alignment is a critical topic in the computer vision community. Numerous efforts have been made and various benchmark datasets have been released in recent decades. However, two significant issues remain in recent datasets, e.g., Intra-Dataset Variation and Inter-Dataset Variation. Inter-Dataset Variation refers to bias on expression, head pose, etc. inside one certain dataset, while Intra-Dataset Variation refers to different bias across different datasets. To address the mentioned problems, we proposed a novel Deep Variation Leveraging Network (DVLN), which consists of two strong coupling sub-networks, e.g., Dataset-Across Network (DA-Net) and Candidate-Decision Network (CD-Net). Extensive evaluations show that our approach demonstrates real-time performance and dramatically outperforms state-of-the-art methods on the challenging 300-W dataset.,,,,,, To address the mentioned problems, we proposed a novel Deep Variation Leveraging Network (DVLN), which consists of two strong coupling sub-networks, e.g., Dataset-Across Network (DA-Net) and Candidate-Decision Network (CD-Net). In particular, DA-Net takes advantage of different characteristics and distributions across different datasets, while CD-Net makes a final decision on candidate hypotheses given by DA-Net to leverage variations within one certain dataset. Extensive evaluations show that our approach demonstrates real-time performance and dramatically outperforms state-of-the-art methods on the challenging 300-W dataset.
{"title":"Leveraging Intra and Inter-Dataset Variations for Robust Face Alignment","authors":"Wenyan Wu, Shuo Yang","doi":"10.1109/CVPRW.2017.261","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.261","url":null,"abstract":"Face alignment is a critical topic in the computer vision community. Numerous efforts have been made and various benchmark datasets have been released in recent decades. However, two significant issues remain in recent datasets, e.g., Intra-Dataset Variation and Inter-Dataset Variation. Inter-Dataset Variation refers to bias on expression, head pose, etc. inside one certain dataset, while Intra-Dataset Variation refers to different bias across different datasets. To address the mentioned problems, we proposed a novel Deep Variation Leveraging Network (DVLN), which consists of two strong coupling sub-networks, e.g., Dataset-Across Network (DA-Net) and Candidate-Decision Network (CD-Net). Extensive evaluations show that our approach demonstrates real-time performance and dramatically outperforms state-of-the-art methods on the challenging 300-W dataset.,,,,,, To address the mentioned problems, we proposed a novel Deep Variation Leveraging Network (DVLN), which consists of two strong coupling sub-networks, e.g., Dataset-Across Network (DA-Net) and Candidate-Decision Network (CD-Net). In particular, DA-Net takes advantage of different characteristics and distributions across different datasets, while CD-Net makes a final decision on candidate hypotheses given by DA-Net to leverage variations within one certain dataset. Extensive evaluations show that our approach demonstrates real-time performance and dramatically outperforms state-of-the-art methods on the challenging 300-W dataset.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"26 1","pages":"2096-2105"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81129729","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}
B. Reddy, Ye-Hoon Kim, Sojung Yun, Chanwon Seo, Junik Jang
Driver’s status is crucial because one of the main reasons for motor vehicular accidents is related to driver’s inattention or drowsiness. Drowsiness detector on a car can reduce numerous accidents. Accidents occur because of a single moment of negligence, thus driver monitoring system which works in real-time is necessary. This detector should be deployable to an embedded device and perform at high accuracy. In this paper, a novel approach towards real-time drowsiness detection based on deep learning which can be implemented on a low cost embedded board and performs with a high accuracy is proposed. Main contribution of our paper is compression of heavy baseline model to a light weight model deployable to an embedded board. Moreover, minimized network structure was designed based on facial landmark input to recognize whether driver is drowsy or not. The proposed model achieved an accuracy of 89.5% on 3-class classification and speed of 14.9 frames per second (FPS) on Jetson TK1.
{"title":"Real-Time Driver Drowsiness Detection for Embedded System Using Model Compression of Deep Neural Networks","authors":"B. Reddy, Ye-Hoon Kim, Sojung Yun, Chanwon Seo, Junik Jang","doi":"10.1109/CVPRW.2017.59","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.59","url":null,"abstract":"Driver’s status is crucial because one of the main reasons for motor vehicular accidents is related to driver’s inattention or drowsiness. Drowsiness detector on a car can reduce numerous accidents. Accidents occur because of a single moment of negligence, thus driver monitoring system which works in real-time is necessary. This detector should be deployable to an embedded device and perform at high accuracy. In this paper, a novel approach towards real-time drowsiness detection based on deep learning which can be implemented on a low cost embedded board and performs with a high accuracy is proposed. Main contribution of our paper is compression of heavy baseline model to a light weight model deployable to an embedded board. Moreover, minimized network structure was designed based on facial landmark input to recognize whether driver is drowsy or not. The proposed model achieved an accuracy of 89.5% on 3-class classification and speed of 14.9 frames per second (FPS) on Jetson TK1.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"16 1","pages":"438-445"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90194478","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}
We present a toolmark dataset created using lock cylinders seized during criminal investigations of break-ins. A total number of 197 cylinders from 48 linked criminal cases were photographed under a comparison microscope used by forensic experts for toolmark comparisons. In order to allow an assessment of the influence of different lighting conditions, all images were captured using a ring light with 11 different lighting settings. Further, matching image regions in the toolmark images were manually annotated. In addition to the annotated toolmark images and the annotation tool, extracted toolmark patches are provided for training and testing to allow a quantitative comparison of the performance of different similarity measures. Finally, results from an evaluation using a publicly available state-of-the-art image descriptor based on deep learning are presented to provide a baseline for future publications.
{"title":"FORMS-Locks: A Dataset for the Evaluation of Similarity Measures for Forensic Toolmark Images","authors":"M. Keglevic, Robert Sablatnig","doi":"10.1109/CVPRW.2017.236","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.236","url":null,"abstract":"We present a toolmark dataset created using lock cylinders seized during criminal investigations of break-ins. A total number of 197 cylinders from 48 linked criminal cases were photographed under a comparison microscope used by forensic experts for toolmark comparisons. In order to allow an assessment of the influence of different lighting conditions, all images were captured using a ring light with 11 different lighting settings. Further, matching image regions in the toolmark images were manually annotated. In addition to the annotated toolmark images and the annotation tool, extracted toolmark patches are provided for training and testing to allow a quantitative comparison of the performance of different similarity measures. Finally, results from an evaluation using a publicly available state-of-the-art image descriptor based on deep learning are presented to provide a baseline for future publications.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"21 1","pages":"1890-1897"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87917495","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}
L. Keselman, J. Woodfill, A. Grunnet-Jepsen, A. Bhowmik
We present a comprehensive overview of the stereoscopic Intel RealSense RGBD imaging systems. We discuss these systems' mode-of-operation, functional behavior and include models of their expected performance, shortcomings, and limitations. We provide information about the systems' optical characteristics, their correlation algorithms, and how these properties can affect different applications, including 3D reconstruction and gesture recognition. Our discussion covers the Intel RealSense R200 and RS400.
{"title":"Intel(R) RealSense(TM) Stereoscopic Depth Cameras","authors":"L. Keselman, J. Woodfill, A. Grunnet-Jepsen, A. Bhowmik","doi":"10.1109/CVPRW.2017.167","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.167","url":null,"abstract":"We present a comprehensive overview of the stereoscopic Intel RealSense RGBD imaging systems. We discuss these systems' mode-of-operation, functional behavior and include models of their expected performance, shortcomings, and limitations. We provide information about the systems' optical characteristics, their correlation algorithms, and how these properties can affect different applications, including 3D reconstruction and gesture recognition. Our discussion covers the Intel RealSense R200 and RS400.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"13 1","pages":"1267-1276"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90313840","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}
Yuchen Fan, Humphrey Shi, Jiahui Yu, Ding Liu, Wei Han, Haichao Yu, Zhangyang Wang, Xinchao Wang, Thomas S. Huang
In this paper, balanced two-stage residual networks (BTSRN) are proposed for single image super-resolution. The deep residual design with constrained depth achieves the optimal balance between the accuracy and the speed for super-resolving images. The experiments show that the balanced two-stage structure, together with our lightweight two-layer PConv residual block design, achieves very promising results when considering both accuracy and speed. We evaluated our models on the New Trends in Image Restoration and Enhancement workshop and challenge on image super-resolution (NTIRE SR 2017). Our final model with only 10 residual blocks ranked among the best ones in terms of not only accuracy (6th among 20 final teams) but also speed (2nd among top 6 teams in terms of accuracy). The source code both for training and evaluation is available in https://github.com/ychfan/sr_ntire2017.
针对单幅图像的超分辨率问题，提出了平衡两级残差网络(BTSRN)。深度约束下的深度残差设计实现了超分辨图像精度和速度的最佳平衡。实验表明，平衡两级结构和轻量化两层PConv残块设计在精度和速度方面都取得了很好的效果。我们在图像恢复和增强的新趋势研讨会和图像超分辨率的挑战(NTIRE SR 2017)上评估了我们的模型。我们的最终模型只有10个剩余块，不仅在准确性(在20个最终团队中排名第6)和速度(在前6个团队中排名第2)方面都名列前茅。培训和评估的源代码可从https://github.com/ychfan/sr_ntire2017获得。
{"title":"Balanced Two-Stage Residual Networks for Image Super-Resolution","authors":"Yuchen Fan, Humphrey Shi, Jiahui Yu, Ding Liu, Wei Han, Haichao Yu, Zhangyang Wang, Xinchao Wang, Thomas S. Huang","doi":"10.1109/CVPRW.2017.154","DOIUrl":"https://doi.org/10.1109/CVPRW.2017.154","url":null,"abstract":"In this paper, balanced two-stage residual networks (BTSRN) are proposed for single image super-resolution. The deep residual design with constrained depth achieves the optimal balance between the accuracy and the speed for super-resolving images. The experiments show that the balanced two-stage structure, together with our lightweight two-layer PConv residual block design, achieves very promising results when considering both accuracy and speed. We evaluated our models on the New Trends in Image Restoration and Enhancement workshop and challenge on image super-resolution (NTIRE SR 2017). Our final model with only 10 residual blocks ranked among the best ones in terms of not only accuracy (6th among 20 final teams) but also speed (2nd among top 6 teams in terms of accuracy). The source code both for training and evaluation is available in https://github.com/ychfan/sr_ntire2017.","PeriodicalId":6668,"journal":{"name":"2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)","volume":"88 1","pages":"1157-1164"},"PeriodicalIF":0.0,"publicationDate":"2017-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78537194","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：