Current state-of-the-art Real-Time Semantic Segmentation Model is still not fast enough. They spend too much time on processing images in a deep CNN to grab the spatial and context information. Somehow, this information may not be so deterministic. In this work, we come up with a multi-image input real-time semantic segmentation model with decoupled label supervision. It can decrease the computational time and keep a relatively high precision of semantic segmentation meanwhile. The novelty of our model lies is picking up the decoupled label supervision to be our loss function and combining it with a multi-branch image processing framework. The edge detection module can not only improve the recognition of the differences between object body and edge but also guarantee the processing procedure of our network to be faster enough. Apart from this, the multi-branch image processing framework is not a burden of running time. Our network is trained on difficult datasets like CamVid and has favourable quality in real-time testing. The mean class IoU of our network is 66.6. It is the highest one among all of the other comparisons.
{"title":"MDNet: A Multi-image Input Real-Time Semantic Segmentation Model with Decoupled Supervision","authors":"Yunze Wu","doi":"10.1145/3471287.3471301","DOIUrl":"https://doi.org/10.1145/3471287.3471301","url":null,"abstract":"Current state-of-the-art Real-Time Semantic Segmentation Model is still not fast enough. They spend too much time on processing images in a deep CNN to grab the spatial and context information. Somehow, this information may not be so deterministic. In this work, we come up with a multi-image input real-time semantic segmentation model with decoupled label supervision. It can decrease the computational time and keep a relatively high precision of semantic segmentation meanwhile. The novelty of our model lies is picking up the decoupled label supervision to be our loss function and combining it with a multi-branch image processing framework. The edge detection module can not only improve the recognition of the differences between object body and edge but also guarantee the processing procedure of our network to be faster enough. Apart from this, the multi-branch image processing framework is not a burden of running time. Our network is trained on difficult datasets like CamVid and has favourable quality in real-time testing. The mean class IoU of our network is 66.6. It is the highest one among all of the other comparisons.","PeriodicalId":306474,"journal":{"name":"2021 the 5th International Conference on Information System and Data Mining","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121145559","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}
Driven by the booming of Internet of Things and 4G/5G communications, an increasingly large number of edge devices, e.g., sensors and cell phones, are continuously producing data service requests, which should be processed in high quality. Recent years have seen a paradigm shift from centralized cloud computing toward edge computing. Edge computing is a distributed computing paradigm that utilizes computing and storage resources of edge devices. Compared with traditional cloud computing, edge computing migrates data computation and storage to the edge devices. Recently many technical breakthroughs have been made in edge computing. This survey reviews existing research on edge computing with a focus on the three mainstream dimensions: resource allocation, data fusion and security. We present specific techniques of the three dimensions and how they can contribute to the improvement of edge computing. Emerging and prospective application fields that would benefit from edge computing are also discussed.
{"title":"A Survey on Mainstream Dimensions of Edge Computing","authors":"Yuanda Wang, Haibo Wang, Shigang Chen, Ye Xia","doi":"10.1145/3471287.3471295","DOIUrl":"https://doi.org/10.1145/3471287.3471295","url":null,"abstract":"Driven by the booming of Internet of Things and 4G/5G communications, an increasingly large number of edge devices, e.g., sensors and cell phones, are continuously producing data service requests, which should be processed in high quality. Recent years have seen a paradigm shift from centralized cloud computing toward edge computing. Edge computing is a distributed computing paradigm that utilizes computing and storage resources of edge devices. Compared with traditional cloud computing, edge computing migrates data computation and storage to the edge devices. Recently many technical breakthroughs have been made in edge computing. This survey reviews existing research on edge computing with a focus on the three mainstream dimensions: resource allocation, data fusion and security. We present specific techniques of the three dimensions and how they can contribute to the improvement of edge computing. Emerging and prospective application fields that would benefit from edge computing are also discussed.","PeriodicalId":306474,"journal":{"name":"2021 the 5th International Conference on Information System and Data Mining","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123915074","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}
∗Object detection is a core process for many image processing applications. Using the YoloV3 deep learning approach to object detection, which is trained on a fixed set of objects, transfer learning is applied to learn the features of novel construction objects. Transfer learning typically requires a large dataset of both images and labels, and labeling image data can take a long time. This paper will introduce several preprocessing pipeline approaches as a means of data amplification and data augmentation to enhance a small dataset using a combination of the following transformations: rotation, scaling, flipping, and grayscale conversion. A construction safety helmet detection model is trained using various experimental data preprocessing pipelines and the results are presented.
{"title":"Automated Dataset Amplification and its Application to Small Dataset Object Detection Transfer Learning","authors":"Muhammad R. Abid, Riley Kiefer","doi":"10.1145/3471287.3471305","DOIUrl":"https://doi.org/10.1145/3471287.3471305","url":null,"abstract":"∗Object detection is a core process for many image processing applications. Using the YoloV3 deep learning approach to object detection, which is trained on a fixed set of objects, transfer learning is applied to learn the features of novel construction objects. Transfer learning typically requires a large dataset of both images and labels, and labeling image data can take a long time. This paper will introduce several preprocessing pipeline approaches as a means of data amplification and data augmentation to enhance a small dataset using a combination of the following transformations: rotation, scaling, flipping, and grayscale conversion. A construction safety helmet detection model is trained using various experimental data preprocessing pipelines and the results are presented.","PeriodicalId":306474,"journal":{"name":"2021 the 5th International Conference on Information System and Data Mining","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129008760","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：