Pub Date : 2016-06-30DOI: 10.1109/CBMI.2016.7500265
Xinchao Li, Peng Xu, Yue Shi, M. Larson, A. Hanjalic
In this paper, we present a subclass-representation approach that predicts the probability of a social image belonging to one particular class. We explore the co-occurrence of user-contributed tags to find subclasses with a strong connection to the top level class. We then project each image onto the resulting subclass space, generating a subclass representation for the image. The advantage of our tag-based subclasses is that they have a chance of being more visually stable and easier to model than top-level classes. Our contribution is to demonstrate that a simple and inexpensive method for generating sub-class representations has the ability to improve classification results in the case of tag classes that are visually highly heterogenous. The approach is evaluated on a set of 1 million photos with 10 top-level classes, from the dataset released by the ACM Multimedia 2013 Yahoo! Large-scale Flickr-tag Image Classification Grand Challenge. Experiments show that the proposed system delivers sound performance for visually diverse classes compared with methods that directly model top classes.
{"title":"Simple tag-based subclass representations for visually-varied image classes","authors":"Xinchao Li, Peng Xu, Yue Shi, M. Larson, A. Hanjalic","doi":"10.1109/CBMI.2016.7500265","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500265","url":null,"abstract":"In this paper, we present a subclass-representation approach that predicts the probability of a social image belonging to one particular class. We explore the co-occurrence of user-contributed tags to find subclasses with a strong connection to the top level class. We then project each image onto the resulting subclass space, generating a subclass representation for the image. The advantage of our tag-based subclasses is that they have a chance of being more visually stable and easier to model than top-level classes. Our contribution is to demonstrate that a simple and inexpensive method for generating sub-class representations has the ability to improve classification results in the case of tag classes that are visually highly heterogenous. The approach is evaluated on a set of 1 million photos with 10 top-level classes, from the dataset released by the ACM Multimedia 2013 Yahoo! Large-scale Flickr-tag Image Classification Grand Challenge. Experiments show that the proposed system delivers sound performance for visually diverse classes compared with methods that directly model top classes.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128881157","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 : 2016-06-30DOI: 10.1109/CBMI.2016.7500277
D. Hauger, M. Schedl
In this demo paper, we present the “Music Tweet Map” interface for browsing music listening events on a global scale. These events have been extracted automatically from a large set of microblogs harvested from Twitter. We showcase the major functionalities offered by the interface, i.e., browsing music by time, specific locations, topic clusters learned from tag information, and music charts. Furthermore, music can be explored via artist similarity. To this end, we present a music similarity measure, based on co-occurrence analysis of items in users' listening histories.
{"title":"Music Tweet Map: A browsing interface to explore the microblogosphere of music","authors":"D. Hauger, M. Schedl","doi":"10.1109/CBMI.2016.7500277","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500277","url":null,"abstract":"In this demo paper, we present the “Music Tweet Map” interface for browsing music listening events on a global scale. These events have been extracted automatically from a large set of microblogs harvested from Twitter. We showcase the major functionalities offered by the interface, i.e., browsing music by time, specific locations, topic clusters learned from tag information, and music charts. Furthermore, music can be explored via artist similarity. To this end, we present a music similarity measure, based on co-occurrence analysis of items in users' listening histories.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124249702","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 : 2016-06-30DOI: 10.1109/CBMI.2016.7500269
C. Pantoja, E. Izquierdo
As the Semantic Web gains popularity, the technical challenges of representing and reasoning with imprecise and uncertain information still remain an outstanding issue. The foundation of the Semantic Web is the assertion of relations between entities, but these relations usually do not carry a degree or level of relationship. Using a simple subject-predicate-object triple we can say that “Alice” (subject) “likes” (predicate) “Rock music” (object), but we can not say that she does so with a confidence or level of 80%. We propose the use of the Transferable Belief Model (TBM) as a way to achieve this. Two contributions are presented in this work: an ontology to represent information in a way which is consistent with the TBM, and a reasoner to assess the knowledge of a given system. Tests show the feasibility of applying this model on large scale Semantic Web information, but further optimisations and tests must be performed.
{"title":"A novel architecture of semantic web reasoner based on transferable belief model","authors":"C. Pantoja, E. Izquierdo","doi":"10.1109/CBMI.2016.7500269","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500269","url":null,"abstract":"As the Semantic Web gains popularity, the technical challenges of representing and reasoning with imprecise and uncertain information still remain an outstanding issue. The foundation of the Semantic Web is the assertion of relations between entities, but these relations usually do not carry a degree or level of relationship. Using a simple subject-predicate-object triple we can say that “Alice” (subject) “likes” (predicate) “Rock music” (object), but we can not say that she does so with a confidence or level of 80%. We propose the use of the Transferable Belief Model (TBM) as a way to achieve this. Two contributions are presented in this work: an ontology to represent information in a way which is consistent with the TBM, and a reasoner to assess the knowledge of a given system. Tests show the feasibility of applying this model on large scale Semantic Web information, but further optimisations and tests must be performed.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125553359","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 : 2016-06-15DOI: 10.1109/CBMI.2016.7500276
A. Moumtzidou, Ilias Gialampoukidis, Theodoros Mironidis, Dimitris Liparas, S. Vrochidis, Y. Kompatsiaris
This paper presents an interactive multimedia search engine, which is capable of searching into multimedia collections by fusing textual and visual information. Apart from multimedia search, the engine is able to perform text search and image retrieval independently using both high-level and low-level information. The images of the multimedia collection are organized by color, offering fast browsing in the image collection.
{"title":"A multimedia interactive search engine based on graph-based and non-linear multimodal fusion","authors":"A. Moumtzidou, Ilias Gialampoukidis, Theodoros Mironidis, Dimitris Liparas, S. Vrochidis, Y. Kompatsiaris","doi":"10.1109/CBMI.2016.7500276","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500276","url":null,"abstract":"This paper presents an interactive multimedia search engine, which is capable of searching into multimedia collections by fusing textual and visual information. Apart from multimedia search, the engine is able to perform text search and image retrieval independently using both high-level and low-level information. The images of the multimedia collection are organized by color, offering fast browsing in the image collection.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124394763","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 : 2016-06-15DOI: 10.1109/CBMI.2016.7500274
R. Licandro, Paolo Rota, M. Reiter, M. Kampel
Acute Lymphoblastic Leukaemia (ALL) is a disease induced by genetic lesion of blood progenitor cells, which influences the hematopoiesis, resulting in the proliferation of undifferentiated (leukaemic) cells. The Minimal Residual Disease (MRD) value is used to quantify these cells and is reliably assessable using Flow CytoMetry (FCM) based measurements. It is a powerful predictor for treatment response and thus used as diagnostic tool for planning patient's individual therapy. In this work we propose an evaluation scheme for longitudinal disease stadium dependent MRD assessment performed on collected clinical data of B-ALL cases after 15, 33 and 78 days of therapy, guided according to the standardised AIEOP-BFM2009 treatment protocol. We compare the blast classification performance using time-specific population models, which are trained using two different core approaches: generative and discriminative. The results show that cell populations change dependent on the observed treatment day and it is identified that a time-specific model of day 15 is not suitable to estimate leukaemic cell populations at treatment day 33 and 78, independent of the methodologies evaluated.
{"title":"Flow Cytometry based automatic MRD assessment in Acute Lymphoblastic Leukaemia: Longitudinal evaluation of time-specific cell population models","authors":"R. Licandro, Paolo Rota, M. Reiter, M. Kampel","doi":"10.1109/CBMI.2016.7500274","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500274","url":null,"abstract":"Acute Lymphoblastic Leukaemia (ALL) is a disease induced by genetic lesion of blood progenitor cells, which influences the hematopoiesis, resulting in the proliferation of undifferentiated (leukaemic) cells. The Minimal Residual Disease (MRD) value is used to quantify these cells and is reliably assessable using Flow CytoMetry (FCM) based measurements. It is a powerful predictor for treatment response and thus used as diagnostic tool for planning patient's individual therapy. In this work we propose an evaluation scheme for longitudinal disease stadium dependent MRD assessment performed on collected clinical data of B-ALL cases after 15, 33 and 78 days of therapy, guided according to the standardised AIEOP-BFM2009 treatment protocol. We compare the blast classification performance using time-specific population models, which are trained using two different core approaches: generative and discriminative. The results show that cell populations change dependent on the observed treatment day and it is identified that a time-specific model of day 15 is not suitable to estimate leukaemic cell populations at treatment day 33 and 78, independent of the methodologies evaluated.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"132 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127210034","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 : 2016-06-15DOI: 10.1109/CBMI.2016.7500278
H. B. D. Silva, Zenilton K. G. Patrocínio, G. Gravier, L. Amsaleg, A. Araújo, S. Guimarães
The huge amount of redundant multimedia data, like video, has become a problem in terms of both space and copyright. Usually, the methods for identifying near-duplicate videos are neither adequate nor scalable to find pairs of similar videos. Similarity self-join operation could be an alternative to solve this problem in which all similar pairs of elements from a video dataset are retrieved. Nonetheless, methods for similarity self-join have poor performance when applied to high-dimensional data. In this work, we propose a new approximate method to compute similarity self-join in sub-quadratic time in order to solve the near-duplicate video detection problem. Our strategy is based on clustering techniques to find out groups of videos which are similar to each other.
{"title":"Near-duplicate video detection based on an approximate similarity self-join strategy","authors":"H. B. D. Silva, Zenilton K. G. Patrocínio, G. Gravier, L. Amsaleg, A. Araújo, S. Guimarães","doi":"10.1109/CBMI.2016.7500278","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500278","url":null,"abstract":"The huge amount of redundant multimedia data, like video, has become a problem in terms of both space and copyright. Usually, the methods for identifying near-duplicate videos are neither adequate nor scalable to find pairs of similar videos. Similarity self-join operation could be an alternative to solve this problem in which all similar pairs of elements from a video dataset are retrieved. Nonetheless, methods for similarity self-join have poor performance when applied to high-dimensional data. In this work, we propose a new approximate method to compute similarity self-join in sub-quadratic time in order to solve the near-duplicate video detection problem. Our strategy is based on clustering techniques to find out groups of videos which are similar to each other.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127795170","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 : 2016-06-15DOI: 10.1109/CBMI.2016.7500255
Alexis Zubiolo, E. Debreuve, D. Ambrosetti, P. Pognonec, X. Descombes
The renal cell carcinoma (RCC) is the most frequent type of kidney cancer (between 90% and 95%). Twelve subtypes of RCC can be distinguished, among which the clear cell carcinoma (ccRCC) and the papillary carcinoma (pRCC) are the two most common ones (75% and 10% of the cases, respectively). After resection (i.e., surgical removal), the tumor is prepared for histological examination (fixation, slicing, staining, observation with a microscope). Along with protein expression and genetic tests, the histological study allows to classify the tumor and define its grade in order to make a prognosis and to take decisions for a potential additional chemotherapy treatment. Digital histology is a recent domain, since routinely, histological slices are studied directly under the microscope. The pioneer works deal with the automatic analysis of cells. However, a crucial factor for RCC classification is the tumoral architecture relying on the structure of the vascular network. For example, coarsely speaking, ccRCC is characterized by a “fishnet” structure while the pRCC has a tree-like structure. To our knowledge, no computerized analysis of the vascular network has been proposed yet. In this context, we developed a complete pipeline to extract the vascular network of a given histological slice and compute features of the underlying graph structure. Then, we studied the potential of such a feature-based approach in classifying a tumor into ccRCC or pRCC. Preliminary results on patient data are encouraging.
{"title":"Is the vascular network discriminant enough to classify renal cell carcinoma?","authors":"Alexis Zubiolo, E. Debreuve, D. Ambrosetti, P. Pognonec, X. Descombes","doi":"10.1109/CBMI.2016.7500255","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500255","url":null,"abstract":"The renal cell carcinoma (RCC) is the most frequent type of kidney cancer (between 90% and 95%). Twelve subtypes of RCC can be distinguished, among which the clear cell carcinoma (ccRCC) and the papillary carcinoma (pRCC) are the two most common ones (75% and 10% of the cases, respectively). After resection (i.e., surgical removal), the tumor is prepared for histological examination (fixation, slicing, staining, observation with a microscope). Along with protein expression and genetic tests, the histological study allows to classify the tumor and define its grade in order to make a prognosis and to take decisions for a potential additional chemotherapy treatment. Digital histology is a recent domain, since routinely, histological slices are studied directly under the microscope. The pioneer works deal with the automatic analysis of cells. However, a crucial factor for RCC classification is the tumoral architecture relying on the structure of the vascular network. For example, coarsely speaking, ccRCC is characterized by a “fishnet” structure while the pRCC has a tree-like structure. To our knowledge, no computerized analysis of the vascular network has been proposed yet. In this context, we developed a complete pipeline to extract the vascular network of a given histological slice and compute features of the underlying graph structure. Then, we studied the potential of such a feature-based approach in classifying a tumor into ccRCC or pRCC. Preliminary results on patient data are encouraging.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"219 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122062635","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 : 2016-06-15DOI: 10.1109/CBMI.2016.7500260
Ionut Cosmin Duta, J. Uijlings, T. Nguyen, K. Aizawa, Alexander Hauptmann, B. Ionescu, N. Sebe
Besides appearance information, the video contains temporal evolution, which represents an important and useful source of information about its content. Many video representation approaches are based on the motion information within the video. The common approach to extract the motion information is to compute the optical flow from the vertical and the horizontal temporal evolution of two consecutive frames. However, the computation of optical flow is very demanding in terms of computational cost, in many cases being the most significant processing step within the overall pipeline of the target video analysis application. In this work we propose a very efficient approach to capture the motion information within the video. Our method is based on a simple temporal and spatial derivation, which captures the changes between two consecutive frames. The proposed descriptor, Histograms of Motion Gradients (HMG), is validated on the UCF50 human action recognition dataset. Our HMG pipeline with several additional speed-ups is able to achieve real-time video processing and outperforms several well-known descriptors including descriptors based on the costly optical flow.
{"title":"Histograms of Motion Gradients for real-time video classification","authors":"Ionut Cosmin Duta, J. Uijlings, T. Nguyen, K. Aizawa, Alexander Hauptmann, B. Ionescu, N. Sebe","doi":"10.1109/CBMI.2016.7500260","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500260","url":null,"abstract":"Besides appearance information, the video contains temporal evolution, which represents an important and useful source of information about its content. Many video representation approaches are based on the motion information within the video. The common approach to extract the motion information is to compute the optical flow from the vertical and the horizontal temporal evolution of two consecutive frames. However, the computation of optical flow is very demanding in terms of computational cost, in many cases being the most significant processing step within the overall pipeline of the target video analysis application. In this work we propose a very efficient approach to capture the motion information within the video. Our method is based on a simple temporal and spatial derivation, which captures the changes between two consecutive frames. The proposed descriptor, Histograms of Motion Gradients (HMG), is validated on the UCF50 human action recognition dataset. Our HMG pipeline with several additional speed-ups is able to achieve real-time video processing and outperforms several well-known descriptors including descriptors based on the costly optical flow.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"79 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126188424","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 : 2016-06-15DOI: 10.1109/CBMI.2016.7500264
Subhasree Basu, Yi Yu, Roger Zimmermann
Lecture videos constitute an important part of the e-learning paradigm. These online video-lectures contain multimedia materials aimed at explaining complex concepts in a more effective way. The videos are mostly grouped by their subjects. However, often there are overlaps between the subjects, e.g. Mathematics and Statistics. Hence, educational content-wise, some of the lecture videos can belong to more than one subject. When they are labeled by only one subject, students searching for the content of the lecture might miss some of these videos. To solve this problem, we aim to provide a clustering of these lecture videos based on their educational content rather than their titles so that such lectures will not be missed out based on the subject labels. Our novel algorithm uses topic modeling on video transcripts generated by automatic captions to extract the contents of these videos. We choose representative text documents for each of the clusters from the Wikipedia. Then we calculate a similarity between the topics extracted from the videos and those of the representative documents of the clusters. Finally we apply fuzzy clustering based on these similarity values and provide a lecture-content based clustering for these lecture videos. The initial results are plausible and confirm the effectiveness of the proposed scheme.
{"title":"Fuzzy clustering of lecture videos based on topic modeling","authors":"Subhasree Basu, Yi Yu, Roger Zimmermann","doi":"10.1109/CBMI.2016.7500264","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500264","url":null,"abstract":"Lecture videos constitute an important part of the e-learning paradigm. These online video-lectures contain multimedia materials aimed at explaining complex concepts in a more effective way. The videos are mostly grouped by their subjects. However, often there are overlaps between the subjects, e.g. Mathematics and Statistics. Hence, educational content-wise, some of the lecture videos can belong to more than one subject. When they are labeled by only one subject, students searching for the content of the lecture might miss some of these videos. To solve this problem, we aim to provide a clustering of these lecture videos based on their educational content rather than their titles so that such lectures will not be missed out based on the subject labels. Our novel algorithm uses topic modeling on video transcripts generated by automatic captions to extract the contents of these videos. We choose representative text documents for each of the clusters from the Wikipedia. Then we calculate a similarity between the topics extracted from the videos and those of the representative documents of the clusters. Finally we apply fuzzy clustering based on these similarity values and provide a lecture-content based clustering for these lecture videos. The initial results are plausible and confirm the effectiveness of the proposed scheme.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131225411","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 : 2016-06-15DOI: 10.1109/CBMI.2016.7500244
Sotiris Karavarsamis, D. Ververidis, G. Chantas, S. Nikolopoulos, Y. Kompatsiaris
In this paper, we explore building classifiers to detect Salsa dance step primitives in choreographies available in the Huawei 3DLife data set. These can collectively be an important component of dance tuition systems that support e-learning. A dance step is reasoned as the shortest possible extract of bodily motion that can uniquely identify a particularly repeatable movement through time. The representation of dance steps adopted is a concatenation of vectorized matrices involving the 3D coordinates of tracked body joints. Under this modeling context, a Salsa dance performance is seen as an ordered sequence of Salsa dance steps, requiring a multiple of the variables allocated in the representation of a single step. Following a previous work by Masurelle & Essid that discusses the classification of six Salsa dance steps from 3DLife, we show that it is possible to obtain better classifiers under a similar experimental protocol in terms of both test accuracy and F-measure. By carefully re-annotating the data in 3DLife, we refocus on the six-step classification problem and then extend the protocol to the case of 20 dance steps. In comparison to common classifiers of the trade operating on full-dimensions, we show that it is possible to produce more accurate models by computing a subspace of the data. At the same time it is possible to reduce problematic bias in resulting models due to the uneven distribution of samples across step data classes. We provide and discuss experimental findings to support both hypotheses for the two experimental settings.
{"title":"Classifying Salsa dance steps from skeletal poses","authors":"Sotiris Karavarsamis, D. Ververidis, G. Chantas, S. Nikolopoulos, Y. Kompatsiaris","doi":"10.1109/CBMI.2016.7500244","DOIUrl":"https://doi.org/10.1109/CBMI.2016.7500244","url":null,"abstract":"In this paper, we explore building classifiers to detect Salsa dance step primitives in choreographies available in the Huawei 3DLife data set. These can collectively be an important component of dance tuition systems that support e-learning. A dance step is reasoned as the shortest possible extract of bodily motion that can uniquely identify a particularly repeatable movement through time. The representation of dance steps adopted is a concatenation of vectorized matrices involving the 3D coordinates of tracked body joints. Under this modeling context, a Salsa dance performance is seen as an ordered sequence of Salsa dance steps, requiring a multiple of the variables allocated in the representation of a single step. Following a previous work by Masurelle & Essid that discusses the classification of six Salsa dance steps from 3DLife, we show that it is possible to obtain better classifiers under a similar experimental protocol in terms of both test accuracy and F-measure. By carefully re-annotating the data in 3DLife, we refocus on the six-step classification problem and then extend the protocol to the case of 20 dance steps. In comparison to common classifiers of the trade operating on full-dimensions, we show that it is possible to produce more accurate models by computing a subspace of the data. At the same time it is possible to reduce problematic bias in resulting models due to the uneven distribution of samples across step data classes. We provide and discuss experimental findings to support both hypotheses for the two experimental settings.","PeriodicalId":356608,"journal":{"name":"2016 14th International Workshop on Content-Based Multimedia Indexing (CBMI)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2016-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124102479","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：