Pub Date : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859539
Ziran Qin, Huanyu He, Weiyao Lin
The high computational cost and huge amount of parameters of 3D Convolutional Neural Networks(CNN) limit the deployment of 3D CNN-based models to lightweight devices with weak computing ability. Applying Winograd layer that combines the idea of pruning and Winograd algorithms to 3D convolution is a promising solution. However, the irregular and unstructured sparsity of 3D Winograd layer makes it difficult to implement efficient compression. In this paper, we propose Regular Mask Pruning to obtain the regular sparse 3D Winograd layer that can be easily compressed. Furthermore, we present Masked 3D Winograd Layer, which can store the compressed parameters of regular sparse Winograd Layer and reduce the number of multiplications in inference. For C3D-based Winograd model, we obtain a regular sparsity of 38.7% without accuracy loss. By our method, the parameters of Winograd layers and the multiplication operations of 3D convolution can be reduced up to 5.32× and 18× respectively.
{"title":"3D Winograd Layer with Regular Mask Pruning","authors":"Ziran Qin, Huanyu He, Weiyao Lin","doi":"10.1109/ICMEW56448.2022.9859539","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859539","url":null,"abstract":"The high computational cost and huge amount of parameters of 3D Convolutional Neural Networks(CNN) limit the deployment of 3D CNN-based models to lightweight devices with weak computing ability. Applying Winograd layer that combines the idea of pruning and Winograd algorithms to 3D convolution is a promising solution. However, the irregular and unstructured sparsity of 3D Winograd layer makes it difficult to implement efficient compression. In this paper, we propose Regular Mask Pruning to obtain the regular sparse 3D Winograd layer that can be easily compressed. Furthermore, we present Masked 3D Winograd Layer, which can store the compressed parameters of regular sparse Winograd Layer and reduce the number of multiplications in inference. For C3D-based Winograd model, we obtain a regular sparsity of 38.7% without accuracy loss. By our method, the parameters of Winograd layers and the multiplication operations of 3D convolution can be reduced up to 5.32× and 18× respectively.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131690031","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859491
Heming Zhao, Yu Liu, Linhui Wei, Yumei Wang
As one of the most popular immersive data formats, point cloud has attracted attention due to its flexibility and simplicity. The advantage of multi-view in light field can capture rich scene information and be effectively applied to 3D point cloud reconstruction. However, existing point cloud reconstruction methods based on light field depth estimation often cause outlier points at the edges of objects, which greatly disturbs the visual effects. In this paper, we propose a superpixel-based optimization scheme for point cloud, which is reconstructed from light field. We use the superpixel-based edge detection algorithm and designed joint bilateral filter to optimize the blurred edge of the depth map, which is obtained from the EPI-based depth estimation. Then, minimal residual outlier points are removed by statistical outlier filter after point cloud generating. Experimental results show that the proposed method increases at least 25.8% in level of details (LoD) compared with several state-of-the-art methods for the real-world and synthetic light field datasets. Besides, the proposed method can restore outlier points reliably and retain the sharp features of point cloud.
{"title":"Superpixel-Based Optimization for Point Cloud Reconstruction from Light Field","authors":"Heming Zhao, Yu Liu, Linhui Wei, Yumei Wang","doi":"10.1109/ICMEW56448.2022.9859491","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859491","url":null,"abstract":"As one of the most popular immersive data formats, point cloud has attracted attention due to its flexibility and simplicity. The advantage of multi-view in light field can capture rich scene information and be effectively applied to 3D point cloud reconstruction. However, existing point cloud reconstruction methods based on light field depth estimation often cause outlier points at the edges of objects, which greatly disturbs the visual effects. In this paper, we propose a superpixel-based optimization scheme for point cloud, which is reconstructed from light field. We use the superpixel-based edge detection algorithm and designed joint bilateral filter to optimize the blurred edge of the depth map, which is obtained from the EPI-based depth estimation. Then, minimal residual outlier points are removed by statistical outlier filter after point cloud generating. Experimental results show that the proposed method increases at least 25.8% in level of details (LoD) compared with several state-of-the-art methods for the real-world and synthetic light field datasets. Besides, the proposed method can restore outlier points reliably and retain the sharp features of point cloud.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"39 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125451298","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859523
Yunqing Wang, Chi-Yo Tsai, Jingning Han, Yaowu Xu
The AV1 video compression format is developed by the Alliance for Open Media (AOMedia) industry consortium, and achieves more than a 30% reduction in bit rate compared to its predecessor VP9 for the same decoded video quality. However, the encoder complexity of AV1 is much higher than VP9. In this paper, we discussed the optimization technologies used to reduce AV1 encoder complexity to the complexity level of VP9 encoder while still achieving 22% bit rate savings. The optimized libaom AV1 encoder offers a superb solution for video-on-demand (VOD) applications, reducing the encoding cost and generating huge bandwidth and storage savings.
{"title":"High Performant AV1 for VOD Applications","authors":"Yunqing Wang, Chi-Yo Tsai, Jingning Han, Yaowu Xu","doi":"10.1109/ICMEW56448.2022.9859523","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859523","url":null,"abstract":"The AV1 video compression format is developed by the Alliance for Open Media (AOMedia) industry consortium, and achieves more than a 30% reduction in bit rate compared to its predecessor VP9 for the same decoded video quality. However, the encoder complexity of AV1 is much higher than VP9. In this paper, we discussed the optimization technologies used to reduce AV1 encoder complexity to the complexity level of VP9 encoder while still achieving 22% bit rate savings. The optimized libaom AV1 encoder offers a superb solution for video-on-demand (VOD) applications, reducing the encoding cost and generating huge bandwidth and storage savings.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116895490","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859337
Kaiyuan Dong, Yuang Zhang, Aixin Zhang
General object detection methods detect various objects independently without paying attention to the relationship between objects. Independent detection of hands and hand-held objects does not take the hand-object relation into account, so the object detector does not achieve optimal performance in such scenes with associated objects. The detection of hand-held objects is an important topic for its wide applications, but few works have been done to model this specific scenario. In this paper, we uses information about the interaction between the hand and the hand-held object as well as causal information when observing the hand-held object as a person to build a novel hand-held object detection model. Experimental results show that our method greatly improves hand-held object detection performance.
{"title":"Interaction Guided Hand-Held Object Detection","authors":"Kaiyuan Dong, Yuang Zhang, Aixin Zhang","doi":"10.1109/ICMEW56448.2022.9859337","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859337","url":null,"abstract":"General object detection methods detect various objects independently without paying attention to the relationship between objects. Independent detection of hands and hand-held objects does not take the hand-object relation into account, so the object detector does not achieve optimal performance in such scenes with associated objects. The detection of hand-held objects is an important topic for its wide applications, but few works have been done to model this specific scenario. In this paper, we uses information about the interaction between the hand and the hand-held object as well as causal information when observing the hand-held object as a person to build a novel hand-held object detection model. Experimental results show that our method greatly improves hand-held object detection performance.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126913543","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859457
Yichen Zhou, Xinfeng Zhang, Shanshe Wang, Lin Li
3D point cloud super-resolution (PCSR) plays an important role in many applications, which can infer a dense geometric shape from a sparse one. However, existing PCSR methods only leverage the geometric properties to predict dense geometric coordinates without considering the importance of correlated attributes in the prediction of complex geometric structures. In this paper, we propose a novel PCSR network by leveraging color attributes to improve the reconstruction quality of dense geometric shape. In the proposed network, we utilize graph convolutions to obtain cross-domain structure representation for point cloud from both geometric coordinates and color attributes, which is constructed by aggregating local points based on the similarity of cross-domain features. Furthermore, we propose a shape-aware loss function to cooperate with network training, which constrains the point cloud generation from both overall and detailed aspects. Extensive experimental results show that our proposed method outperforms the state-of-the-art methods from both objective and subjective quality.
{"title":"Multi-Attribute Joint Point Cloud Super-Resolution with Adversarial Feature Graph Networks","authors":"Yichen Zhou, Xinfeng Zhang, Shanshe Wang, Lin Li","doi":"10.1109/ICMEW56448.2022.9859457","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859457","url":null,"abstract":"3D point cloud super-resolution (PCSR) plays an important role in many applications, which can infer a dense geometric shape from a sparse one. However, existing PCSR methods only leverage the geometric properties to predict dense geometric coordinates without considering the importance of correlated attributes in the prediction of complex geometric structures. In this paper, we propose a novel PCSR network by leveraging color attributes to improve the reconstruction quality of dense geometric shape. In the proposed network, we utilize graph convolutions to obtain cross-domain structure representation for point cloud from both geometric coordinates and color attributes, which is constructed by aggregating local points based on the similarity of cross-domain features. Furthermore, we propose a shape-aware loss function to cooperate with network training, which constrains the point cloud generation from both overall and detailed aspects. Extensive experimental results show that our proposed method outperforms the state-of-the-art methods from both objective and subjective quality.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"55 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128549045","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859373
Ekrem Çetinkaya, Hadi Amirpour, C. Timmerer
Light field imaging enables post-capture actions such as refocusing and changing view perspective by capturing both spatial and angular information. However, capturing richer information of the 3D scene results in a huge amount of data. To improve the compression efficiency of the existing light field compression methods, we investigate the impact of light field super-resolution approaches (both spatial and angular super-resolution) on the compression efficiency. To this end, firstly, we downscale light field images over (i) spatial resolution, (ii) angular resolution, and (iii) spatial-angular resolution and encode them using Versatile Video Coding (VVC). We then apply a set of light field super-resolution deep neural networks to reconstruct light field images in their full spatial-angular resolution and compare their compression efficiency. Experimental results show that encoding the low angular resolution light field image and applying angular super-resolution yield bitrate savings of 51.16% and 53.41% to maintain the same PSNR and SSIM, respectively, compared to encoding the light field image in high-resolution.
{"title":"LFC-SASR: Light Field Coding Using Spatial and Angular Super-Resolution","authors":"Ekrem Çetinkaya, Hadi Amirpour, C. Timmerer","doi":"10.1109/ICMEW56448.2022.9859373","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859373","url":null,"abstract":"Light field imaging enables post-capture actions such as refocusing and changing view perspective by capturing both spatial and angular information. However, capturing richer information of the 3D scene results in a huge amount of data. To improve the compression efficiency of the existing light field compression methods, we investigate the impact of light field super-resolution approaches (both spatial and angular super-resolution) on the compression efficiency. To this end, firstly, we downscale light field images over (i) spatial resolution, (ii) angular resolution, and (iii) spatial-angular resolution and encode them using Versatile Video Coding (VVC). We then apply a set of light field super-resolution deep neural networks to reconstruct light field images in their full spatial-angular resolution and compare their compression efficiency. Experimental results show that encoding the low angular resolution light field image and applying angular super-resolution yield bitrate savings of 51.16% and 53.41% to maintain the same PSNR and SSIM, respectively, compared to encoding the light field image in high-resolution.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"166 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132756321","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859437
Hannes Fassold
The automatic detection and tracking of objects in a video is crucial for many video understanding tasks. We propose a novel deep learning based algorithm for object detection and tracking, which is able to detect more than 1,000 object classes and tracks them robustly, even for challenging content. The robustness of the tracking is due to the usage of optical flow information. Additionally, we utilize only the part of the bounding box corresponding to the object shape for the tracking.
{"title":"Detic-Track: Robust Detection and Tracking of Objects in Video","authors":"Hannes Fassold","doi":"10.1109/ICMEW56448.2022.9859437","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859437","url":null,"abstract":"The automatic detection and tracking of objects in a video is crucial for many video understanding tasks. We propose a novel deep learning based algorithm for object detection and tracking, which is able to detect more than 1,000 object classes and tracks them robustly, even for challenging content. The robustness of the tracking is due to the usage of optical flow information. Additionally, we utilize only the part of the bounding box corresponding to the object shape for the tracking.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"76 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124104539","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859411
Chih-Chang Yu, Chih-Ching Chang, Hsu-Yung Cheng
Information collection and analysis have played a very important role in high-level baseball competitions. Knowing opponent’s possible strategies or weakness can help own team plan adequate countermeasures. The purpose of this study is to explore how artificial intelligence technology can be applied to this domain. This study focuses on the pitching events in baseball. The goal is to predict the pitch types that a pitcher may throw in the next pitch according to the situation on the field. To achieve this, we mine discriminative features from baseball statistics and propose a stacked long-term and short-term memory model (LSTM) with attention mechanism. Experimental data come from the pitching data of 201 pitchers in Major League Baseball from 2016 to 2021. By collecting information of pitchers’ pitching statistics and on-field situations, results show that the average accuracy rate reaches 76.7%, outperforming conventional machine learning prediction models.
{"title":"Decide the Next Pitch: A Pitch Prediction Model Using Attention-Based LSTM","authors":"Chih-Chang Yu, Chih-Ching Chang, Hsu-Yung Cheng","doi":"10.1109/ICMEW56448.2022.9859411","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859411","url":null,"abstract":"Information collection and analysis have played a very important role in high-level baseball competitions. Knowing opponent’s possible strategies or weakness can help own team plan adequate countermeasures. The purpose of this study is to explore how artificial intelligence technology can be applied to this domain. This study focuses on the pitching events in baseball. The goal is to predict the pitch types that a pitcher may throw in the next pitch according to the situation on the field. To achieve this, we mine discriminative features from baseball statistics and propose a stacked long-term and short-term memory model (LSTM) with attention mechanism. Experimental data come from the pitching data of 201 pitchers in Major League Baseball from 2016 to 2021. By collecting information of pitchers’ pitching statistics and on-field situations, results show that the average accuracy rate reaches 76.7%, outperforming conventional machine learning prediction models.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"96 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122532050","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859503
Hsiu-Chieh Lee, Che-Hsien Lin, Li-Chen Cheng, S. Hsu, Jun-Cheng Chen, Chih-Yu Wang
Kinship face synthesis has been drawing significant attention recently. Nevertheless, it is challenging to evaluate the performance of the generated images due to lacking ground truths and objective metrics. In this paper, we propose an interactive online website for kinship synthesis model evaluation. The website not only showcases the result of the model but also collects feedback data from the users for analysis and model improvement. The website allows users to 1) build their kinship dataset and generate kinship images with our models, and 2) play games to distinguish between a real child face image and synthesized child images. Through the incentives triggered by the gamified rating mechanism, we expect the collected feedback data will be more promising.
{"title":"Kinship Face Synthesis Evaluation Website with Gamified Mechanism","authors":"Hsiu-Chieh Lee, Che-Hsien Lin, Li-Chen Cheng, S. Hsu, Jun-Cheng Chen, Chih-Yu Wang","doi":"10.1109/ICMEW56448.2022.9859503","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859503","url":null,"abstract":"Kinship face synthesis has been drawing significant attention recently. Nevertheless, it is challenging to evaluate the performance of the generated images due to lacking ground truths and objective metrics. In this paper, we propose an interactive online website for kinship synthesis model evaluation. The website not only showcases the result of the model but also collects feedback data from the users for analysis and model improvement. The website allows users to 1) build their kinship dataset and generate kinship images with our models, and 2) play games to distinguish between a real child face image and synthesized child images. Through the incentives triggered by the gamified rating mechanism, we expect the collected feedback data will be more promising.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116616657","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 : 2022-07-18DOI: 10.1109/ICMEW56448.2022.9859471
Hang Wu, Xi Chen, Xuelong Li, Haokai Ma, Yuze Zheng, Xiangxian Li, Xiangxu Meng, Lei Meng
This demo illustrates a visually-aware food analysis (VAFA) system for socially-engaged diet management. VAFA is able to receive multimedia inputs, such as the images of food with/without a description to record a user’s daily diet. Such information will be passed to AI algorithms for food classification, ingredient recognition, and nutrition analysis, to produce a nutrition report for the user. Moreover, VAFA profiles the users’ eating habits to make personalized recipe recommendation and identify the social communities with similar eating preferences. VAFA is empowered by state-of-the-art AI algorithms and a large-scale dataset with 300K users, 400K recipes, and over 10M user-recipe interactions.
{"title":"A Visually-Aware Food Analysis System for Diet Management","authors":"Hang Wu, Xi Chen, Xuelong Li, Haokai Ma, Yuze Zheng, Xiangxian Li, Xiangxu Meng, Lei Meng","doi":"10.1109/ICMEW56448.2022.9859471","DOIUrl":"https://doi.org/10.1109/ICMEW56448.2022.9859471","url":null,"abstract":"This demo illustrates a visually-aware food analysis (VAFA) system for socially-engaged diet management. VAFA is able to receive multimedia inputs, such as the images of food with/without a description to record a user’s daily diet. Such information will be passed to AI algorithms for food classification, ingredient recognition, and nutrition analysis, to produce a nutrition report for the user. Moreover, VAFA profiles the users’ eating habits to make personalized recipe recommendation and identify the social communities with similar eating preferences. VAFA is empowered by state-of-the-art AI algorithms and a large-scale dataset with 300K users, 400K recipes, and over 10M user-recipe interactions.","PeriodicalId":106759,"journal":{"name":"2022 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128948195","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：