It is widely-known that melanoma is one of the deadliest skin cancers with a very high mortality rate, while it is curable with early identification. Therefore, early detection of melanoma is extremely necessary for the treatment of this disease. In recent decades, Convolutional Neural Networks (CNN) have achieved state-of-the-art performance in many different visual classification tasks, so they have also been employed in melanoma recognition tasks. Due to the complexity of the deep learning model and huge numbers of parameters, a large amount of labelled data is required to achieve a better training performance. However, in practical settings, it is difficult for many applications to obtain enough labelled sample data. This paper explore to solve this problems based on data augmentation strategy. In the experiment conducted in our paper, the training data is augmented through CycleGAN-based approaches to generate more training samples with detailed information, and then the CNN model can be trained using the artificially enlarged dataset. The experimental results show that the combination of CycleGAN data augmentation method and EfficientNet B1 can effectively saves the cost of manual annotation, while dramatically improves classification accuracy.
{"title":"CycleGAN Based Data Augmentation For Melanoma images Classification","authors":"Yixin Chen, Yifan Zhu, Yanfeng Chang","doi":"10.1145/3430199.3430217","DOIUrl":"https://doi.org/10.1145/3430199.3430217","url":null,"abstract":"It is widely-known that melanoma is one of the deadliest skin cancers with a very high mortality rate, while it is curable with early identification. Therefore, early detection of melanoma is extremely necessary for the treatment of this disease. In recent decades, Convolutional Neural Networks (CNN) have achieved state-of-the-art performance in many different visual classification tasks, so they have also been employed in melanoma recognition tasks. Due to the complexity of the deep learning model and huge numbers of parameters, a large amount of labelled data is required to achieve a better training performance. However, in practical settings, it is difficult for many applications to obtain enough labelled sample data. This paper explore to solve this problems based on data augmentation strategy. In the experiment conducted in our paper, the training data is augmented through CycleGAN-based approaches to generate more training samples with detailed information, and then the CNN model can be trained using the artificially enlarged dataset. The experimental results show that the combination of CycleGAN data augmentation method and EfficientNet B1 can effectively saves the cost of manual annotation, while dramatically improves classification accuracy.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"176 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132381248","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}
Lightweight models, high accuracy and real-time performance are essential for facial landmark detection algorithms. Considering these three aspects, this paper proposes a real-time and efficient face landmark algorithm. First, mobilenetV2 is used as the backbone network. Next, the traditional convolution operation is replaced with deeply separable convolution, and the shallow and deep feature maps are merged to enhance the context connection. Then multi-scale fusion output is used in the output layer to enhance the detection efficiency of small-sized faces. Finally, the Euler angle weights are introduced into the loss function, and the 14 key points in the average face model are compared with the predicted key points. During the training process, this paper proposes rotated the 300W and AFLW datasets in multi-angle to occlude the dataset and enhance the generalization ability of the model. The experimental results show that the proposed algorithm in this paper can achieve real-time and efficient facial landmark detection.
{"title":"Real-time Efficient Facial Landmark Detection Algorithms","authors":"Hanying Xiong, Tongwei Lu, Hongzhi Zhang","doi":"10.1145/3430199.3430200","DOIUrl":"https://doi.org/10.1145/3430199.3430200","url":null,"abstract":"Lightweight models, high accuracy and real-time performance are essential for facial landmark detection algorithms. Considering these three aspects, this paper proposes a real-time and efficient face landmark algorithm. First, mobilenetV2 is used as the backbone network. Next, the traditional convolution operation is replaced with deeply separable convolution, and the shallow and deep feature maps are merged to enhance the context connection. Then multi-scale fusion output is used in the output layer to enhance the detection efficiency of small-sized faces. Finally, the Euler angle weights are introduced into the loss function, and the 14 key points in the average face model are compared with the predicted key points. During the training process, this paper proposes rotated the 300W and AFLW datasets in multi-angle to occlude the dataset and enhance the generalization ability of the model. The experimental results show that the proposed algorithm in this paper can achieve real-time and efficient facial landmark detection.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"74 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127363225","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}
Melanoma has been regarded as one of the fatal skin cancer diseases all around the world. Early detection on melanoma can be quite helpful in the clinical treatment, to prevent the deterioration of the deadly diseases. Handcrafted-feature extraction and shallow architecture-based classifier (such as k-nearest neighbors algorithm, random forest, support vector machine) worked as the basis of the previous attempts in detecting process. During the recent years, the new approach named deep convolutional neural network (CNN) was used for the detecting task. Although the persistent progress and efforts have been achieved, the classification methods desire to go a further step in pursuing further improvement on its performance. The goal of this paper is to improve the detection performance using an ensemble learning framework. Both the personal information (such as the age, gender information of the patients) and latest deep learning approaches are applied in this paper. The two approaches have provided the mutual complements for each other, which demonstrated enormous advantages for the ensemble learning framework in detecting task. We conducted extensive experiments that provide a large dataset for detecting melanoma, which illustrates that our ensemble learning can provide superior performance with high accuracy.
{"title":"Ensembling Learning Based Melanoma Classification Using Gradient Boosting Decision Trees","authors":"Yipeng Han, Xiaolu Zheng","doi":"10.1145/3430199.3430215","DOIUrl":"https://doi.org/10.1145/3430199.3430215","url":null,"abstract":"Melanoma has been regarded as one of the fatal skin cancer diseases all around the world. Early detection on melanoma can be quite helpful in the clinical treatment, to prevent the deterioration of the deadly diseases. Handcrafted-feature extraction and shallow architecture-based classifier (such as k-nearest neighbors algorithm, random forest, support vector machine) worked as the basis of the previous attempts in detecting process. During the recent years, the new approach named deep convolutional neural network (CNN) was used for the detecting task. Although the persistent progress and efforts have been achieved, the classification methods desire to go a further step in pursuing further improvement on its performance. The goal of this paper is to improve the detection performance using an ensemble learning framework. Both the personal information (such as the age, gender information of the patients) and latest deep learning approaches are applied in this paper. The two approaches have provided the mutual complements for each other, which demonstrated enormous advantages for the ensemble learning framework in detecting task. We conducted extensive experiments that provide a large dataset for detecting melanoma, which illustrates that our ensemble learning can provide superior performance with high accuracy.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"102 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123648050","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}
James Pope, Mark G. Terwilliger, J. A. Connell, Gabriel Talley, Nicholas Blozik, David Taylor
The aircraft cargo industry still maintains vast amounts of the maintenance history of aircraft components in electronic (i.e. scanned) but unsearchable images. For a given supplier, there can be hundreds of thousands of image documents only some of which contain useful information. Using supervised machine learning techniques has been shown to be effective in recognising these documents for further information extraction. A well known deficiency of supervised learning approaches is that annotating sufficient documents to create an effective model requires valuable human effort. This paper first shows how to obtain a representative sample from a supplier's corpus. Given this sample of unlabelled documents an active learning approach is used to select which documents to annotate first using a normalised certainty measure derived from a soft classifier's prediction distribution. Finally the accuracy of various selection approaches using this certainty measure are compared along each iteration of the active learning cycle. The experiments show that a greedy selection method using the uncertainty measure can significantly reduce the number of annotations required for a certain accuracy. The results provide valuable information for users and more generally illustrate an effective deployment of a machine learning application.
{"title":"Annotating Documents using Active Learning Methods for a Maintenance Analysis Application","authors":"James Pope, Mark G. Terwilliger, J. A. Connell, Gabriel Talley, Nicholas Blozik, David Taylor","doi":"10.1145/3430199.3430214","DOIUrl":"https://doi.org/10.1145/3430199.3430214","url":null,"abstract":"The aircraft cargo industry still maintains vast amounts of the maintenance history of aircraft components in electronic (i.e. scanned) but unsearchable images. For a given supplier, there can be hundreds of thousands of image documents only some of which contain useful information. Using supervised machine learning techniques has been shown to be effective in recognising these documents for further information extraction. A well known deficiency of supervised learning approaches is that annotating sufficient documents to create an effective model requires valuable human effort. This paper first shows how to obtain a representative sample from a supplier's corpus. Given this sample of unlabelled documents an active learning approach is used to select which documents to annotate first using a normalised certainty measure derived from a soft classifier's prediction distribution. Finally the accuracy of various selection approaches using this certainty measure are compared along each iteration of the active learning cycle. The experiments show that a greedy selection method using the uncertainty measure can significantly reduce the number of annotations required for a certain accuracy. The results provide valuable information for users and more generally illustrate an effective deployment of a machine learning application.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"67 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125217533","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}
In this paper, a non-contact, unmarked computer vision measurement method is presented and applied to measure the two-dimensional (2D) vibration displacement of hoisting vertical ropes. In this method, the primary work is to perform camera calibration of monocular vision using a neural network (NN) model. Then, in the image sequence, a straight line perpendicular to the hoisting rope is added by digital image processing (DIP) method, and their intersection region is regarded as the measuring target. Digital image correlation (DIC) algorithm at sub-pixel level is applied to locate the measuring target in image sequence. This method is used to measure the vibration displacement of an actual hoisting rope in mine, and the measurement results of three targets on the rope are consistent with tiny amplitude differences, which indicates that this method is feasible for the vibration measurement of hoisting vertical rope.
{"title":"Vision-based 2D Vibration Displacement Measurement of Hoisting Vertical Rope in Mine Hoist","authors":"Ganggang Wu, Xingming Xiao, Chi Ma","doi":"10.1145/3430199.3430221","DOIUrl":"https://doi.org/10.1145/3430199.3430221","url":null,"abstract":"In this paper, a non-contact, unmarked computer vision measurement method is presented and applied to measure the two-dimensional (2D) vibration displacement of hoisting vertical ropes. In this method, the primary work is to perform camera calibration of monocular vision using a neural network (NN) model. Then, in the image sequence, a straight line perpendicular to the hoisting rope is added by digital image processing (DIP) method, and their intersection region is regarded as the measuring target. Digital image correlation (DIC) algorithm at sub-pixel level is applied to locate the measuring target in image sequence. This method is used to measure the vibration displacement of an actual hoisting rope in mine, and the measurement results of three targets on the rope are consistent with tiny amplitude differences, which indicates that this method is feasible for the vibration measurement of hoisting vertical rope.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"54 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129071276","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}
In this paper a color multi-secret visual cryptography scheme specifically for (3, 4, 5) access structure is proposed with random colors and XOR operation being leveraged to generate the sharing images. The recovery images with size invariant are obtained by the XOR operation of specific combination of shares. In order to achieve ideal perceptual quality, we present the optimization algorithm with which the visual quality of recovery images is improved significantly without sacrificing computation complexity. Experimental results demonstrate the effectiveness of the proposed scheme.
{"title":"A Novel Color Multi-Secret Visual Cryptography Scheme","authors":"Rui Sun, Zhengxin Fu, Bin Yu, Hangying Huang","doi":"10.1145/3430199.3430235","DOIUrl":"https://doi.org/10.1145/3430199.3430235","url":null,"abstract":"In this paper a color multi-secret visual cryptography scheme specifically for (3, 4, 5) access structure is proposed with random colors and XOR operation being leveraged to generate the sharing images. The recovery images with size invariant are obtained by the XOR operation of specific combination of shares. In order to achieve ideal perceptual quality, we present the optimization algorithm with which the visual quality of recovery images is improved significantly without sacrificing computation complexity. Experimental results demonstrate the effectiveness of the proposed scheme.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126831177","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}
Traditional remote sensing image retrieval (RSIR) methods are generally based on images from a specific single source. As different sources and huge volumes of remote sensing images have been easily available nowadays, RSIR is facing the challenge of retrieving remote sensing images with different spectral and spatial information from different sources. Benefited from compelling image feature extraction ability of deep neural networks and efficient computing power and effective retrieval ability of hashing, deep hashing networks has become prevalent for image retrieval researches. In this paper, a deep hashing network based on split channels is proposed for hybrid source RSIR called split-channels triplet deep hashing networks(SCTDHNs). It takes skillfully splitting channels as input, and is mainly composed of a hybrid source deep hashing subnetwork for cross source images retrieval and single-source deep hashing sub-network for a multi-spectral image retrieval, and each of them achieves high retrieval performance. Furthermore, a novel trick for loss function is proposed, called increased intervals between dissimilar pairs during training stage that dramatically improves the retrieval performance. Extensive experiments implement on dual-source remote sensing data set demonstrate that proposed method yields better performance than existing state-of-art hybrid source retrieval methods as far as is known.
{"title":"Deep Hashing Network Based on Split Channels for Hybrid-Source Remote Sensing Image Retrieval","authors":"Salayidin Sirajidin, H. Huo, T. Fang","doi":"10.1145/3430199.3430225","DOIUrl":"https://doi.org/10.1145/3430199.3430225","url":null,"abstract":"Traditional remote sensing image retrieval (RSIR) methods are generally based on images from a specific single source. As different sources and huge volumes of remote sensing images have been easily available nowadays, RSIR is facing the challenge of retrieving remote sensing images with different spectral and spatial information from different sources. Benefited from compelling image feature extraction ability of deep neural networks and efficient computing power and effective retrieval ability of hashing, deep hashing networks has become prevalent for image retrieval researches. In this paper, a deep hashing network based on split channels is proposed for hybrid source RSIR called split-channels triplet deep hashing networks(SCTDHNs). It takes skillfully splitting channels as input, and is mainly composed of a hybrid source deep hashing subnetwork for cross source images retrieval and single-source deep hashing sub-network for a multi-spectral image retrieval, and each of them achieves high retrieval performance. Furthermore, a novel trick for loss function is proposed, called increased intervals between dissimilar pairs during training stage that dramatically improves the retrieval performance. Extensive experiments implement on dual-source remote sensing data set demonstrate that proposed method yields better performance than existing state-of-art hybrid source retrieval methods as far as is known.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128071862","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}
Some common diseases (such as Parkinson's disease, stroke and epilepsy) could cause spontaneous tremors in patients, and doctors could make a preliminary diagnosis based on these tremor in different parts of the patient's body. In order to be more accurate to automatically obtain the tremor signal, we proposed a Novel method for extracting subtle tremor signal from human body. The scope of traditional video tremor extraction usually contained the whole video. In order to extract tremor signals of different body parts of patients separately, we adopted OpenPose to automatically divide different body parts, so as to obtain more detailed video of body parts. Due to some patients' tremor was not obvious, so we used Eulerian video magnification method to amplify the non-obvious tremor and then extracted the tremor signal from the amplified video. To obtain a better tremor signal, we used Butterworth band-pass filter to remove the noise from the initial signal. The experimental results showed that our method can automatically obtain the tremor signal of different body parts of the patient, and the tremor signal was relatively accurate.
{"title":"A Novel Method for Extracting Subtle Tremor Signal from Human Body","authors":"Weiping Liu, Zhiyang Lin, Guannan Chen","doi":"10.1145/3430199.3430205","DOIUrl":"https://doi.org/10.1145/3430199.3430205","url":null,"abstract":"Some common diseases (such as Parkinson's disease, stroke and epilepsy) could cause spontaneous tremors in patients, and doctors could make a preliminary diagnosis based on these tremor in different parts of the patient's body. In order to be more accurate to automatically obtain the tremor signal, we proposed a Novel method for extracting subtle tremor signal from human body. The scope of traditional video tremor extraction usually contained the whole video. In order to extract tremor signals of different body parts of patients separately, we adopted OpenPose to automatically divide different body parts, so as to obtain more detailed video of body parts. Due to some patients' tremor was not obvious, so we used Eulerian video magnification method to amplify the non-obvious tremor and then extracted the tremor signal from the amplified video. To obtain a better tremor signal, we used Butterworth band-pass filter to remove the noise from the initial signal. The experimental results showed that our method can automatically obtain the tremor signal of different body parts of the patient, and the tremor signal was relatively accurate.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133258380","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}
Today, the increasing ease of publishing information online combined with a gradual shift of paradigm from consuming news via conventional media to non-conventional media calls for a computational and automatic approach to the identification of an article's legitimacy. In this study, we propose an approach for cross-domain fake news detection focusing on the identification of legitimate content from a pool of articles that are of varying degrees of legitimacy. We present a model as a proof of concept as well as data gathered from evaluating the model on Fake-News AMT, a dataset released for cross-domain fake news detection. The results of our model are then compared against a baseline model which has served as the benchmark for the dataset. We find all results in support of our hypothesis. Our proof-of-concept model has also outperformed the benchmark in the domains Technology and Entertainment as well as when it was run on the whole dataset at once.
{"title":"Automatic Differentiation Between Legitimate and Fake News Using Named Entity Recognition","authors":"Bo Xu, C. Tsai","doi":"10.1145/3430199.3430220","DOIUrl":"https://doi.org/10.1145/3430199.3430220","url":null,"abstract":"Today, the increasing ease of publishing information online combined with a gradual shift of paradigm from consuming news via conventional media to non-conventional media calls for a computational and automatic approach to the identification of an article's legitimacy. In this study, we propose an approach for cross-domain fake news detection focusing on the identification of legitimate content from a pool of articles that are of varying degrees of legitimacy. We present a model as a proof of concept as well as data gathered from evaluating the model on Fake-News AMT, a dataset released for cross-domain fake news detection. The results of our model are then compared against a baseline model which has served as the benchmark for the dataset. We find all results in support of our hypothesis. Our proof-of-concept model has also outperformed the benchmark in the domains Technology and Entertainment as well as when it was run on the whole dataset at once.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134091726","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}
On-line Precision scalability of the deep neural networks(DNNs) is a critical feature to support accuracy and complexity trade-off during the DNN inference. In this paper, we propose dual-precision DNN that includes two different precision modes in a single model, thereby supporting an on-line precision switch without re-training. The proposed two-phase training process optimizes both low- and high-precision modes.
{"title":"Dual-Precision Deep Neural Network","authors":"J. Park, J. Choi, J. Ko","doi":"10.1145/3430199.3430228","DOIUrl":"https://doi.org/10.1145/3430199.3430228","url":null,"abstract":"On-line Precision scalability of the deep neural networks(DNNs) is a critical feature to support accuracy and complexity trade-off during the DNN inference. In this paper, we propose dual-precision DNN that includes two different precision modes in a single model, thereby supporting an on-line precision switch without re-training. The proposed two-phase training process optimizes both low- and high-precision modes.","PeriodicalId":371055,"journal":{"name":"Proceedings of the 2020 3rd International Conference on Artificial Intelligence and Pattern Recognition","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-06-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134120222","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：