By using one-bit binary patterns instead of eight-bit sinusoidal ones, the binary defocusing tech... more By using one-bit binary patterns instead of eight-bit sinusoidal ones, the binary defocusing techniques have been widely applied for high-speed 3D shape measurement. As projector defocusing is required, the phase unwrapping process of these techniques remains challenging. A recently proposed ternary Gray-code method can effectively increase the measurement speed by reducing the number of acquired coding patterns. However, it still has limitations, including the measuring range and a noise problem. To improve these, a new ternary Gray-code method is proposed by utilizing 2D modulation instead of the 1D modulation used in the conventional method. Simulations were conducted to compare these two methods, and the results verify that the proposed method reduces the variance in the introduced intermediate gray value by nearly 90%. Experiments were also carried out with the results verifying the superiority of the proposed method.
The vehicle dynamics model has multiple degrees of freedom, with strong nonlinear characteristics... more The vehicle dynamics model has multiple degrees of freedom, with strong nonlinear characteristics, so it is difficult to quickly obtain the accurate target oil pressure of an electronically assisted brake system based on the model. This paper proposes a target oil pressure recognition algorithm based on the T-S fuzzy neural network model. Firstly, the braking conditions classification algorithm is built according to the sampled braking intention data. The data are divided into the emergency braking condition data and the general braking condition data by the braking conditions classification algorithm. Secondly, the recognition model is trained respectively by the different braking condition data sets. In the training process, the fuzzy C-means clustering algorithm is used to identify the antecedent parameters of the model, and the learning rate cosine attenuation strategy is applied to optimize the parameter learning process. Finally, a correction method of target oil pressure base...
This paper presents an SVM--based algorithm for the transfer of knowledge across robot platforms ... more This paper presents an SVM--based algorithm for the transfer of knowledge across robot platforms aiming to perform the same task. Our method exploits efficiently the transferred knowledge while updating incrementally the internal representation as new information is available. The algorithm is adaptive and tends to privilege new data when building the SV solution. This prevents the old knowledge to nest into the model and eventually become a possible source of misleading information. We tested our approach in the domain of vision-based place recognition. Extensive experiments show that using transferred knowledge clearly pays off in terms of performance and stability of the solution.
ELCVIA Electronic Letters on Computer Vision and Image Analysis, 2010
Local features have repeatedly shown their effectiveness for object recognition during the last y... more Local features have repeatedly shown their effectiveness for object recognition during the last years, and they have consequently become the preferred descriptor for this type of problems. The solution of the correspondence problem is traditionally approached with exact or approximate techniques. In this paper we are interested in methods that solve the correspondence problem via the definition of a kernel function that makes it possible to use local features as input to a support vector machine. We single out the match kernel, an exact approach, and the pyramid match kernel, that uses instead an approximate strategy. We present a thorough experimental evaluation of the two methods on three different databases. Results show that the exact method performs consistently better than the approximate one, especially for the object identification task, when training on a decreasing number of images. Based on these findings and on the computational cost of each approach, we suggest some criteria for choosing between the two kernels given the application at hand.
The accurate air-fuel ratio (AFR) control is crucial for the exhaust emission reduction based on ... more The accurate air-fuel ratio (AFR) control is crucial for the exhaust emission reduction based on the three-way catalytic converter in the spark ignition (SI) engine. The difficulties in transient cylinder air mass flow measurement, the existing fuel mass wall-wetting phenomenon, and the unfixed AFR path dynamic variations make the design of the AFR controller a challenging task. In this paper, an adaptive AFR regulation controller is designed using the feedforward and feedback control scheme based on the dynamical modelling of the AFR path. The generalized predictive control method is proposed to solve the problems of inherent nonlinearities, time delays, parameter variations, and uncertainties in the AFR closed loop. The simulation analysis is investigated for the effectiveness of noise suppression, online prediction, and self-correction on the SI engine system. Moreover, the experimental verification shows an acceptable performance of the designed controller and the potential usag...
Fringe projection technologies have been widely used for three-dimensional (3D) shape measurement... more Fringe projection technologies have been widely used for three-dimensional (3D) shape measurement. One of the critical issues is absolute phase recovery, especially for measuring multiple isolated objects. This paper proposes a method for absolute phase retrieval using only one coded pattern. A total of four patterns including one coded pattern and three phase-shift patterns are projected, captured, and processed. The wrapped phase, as well as average intensity and intensity modulation, are calculated from three phase-shift patterns. A code word encrypted into the coded pattern can be calculated using the average intensity and intensity modulation. Based on geometric constraints of fringe projection system, the minimum fringe order map can be created, upon which the fringe order can be calculated from the code word. Compared with the conventional method, the measurement depth range is significantly improved. Finally, the wrapped phase can be unwrapped for absolute phase map. Since o...
Fringe projection systems have been widely applied in three-dimensional (3D) shape measurements. ... more Fringe projection systems have been widely applied in three-dimensional (3D) shape measurements. One of the important issues is how to retrieve the absolute phase. This paper presents a modified gray-level coding method for absolute phase retrieval. Specifically, two groups of fringe patterns are projected onto the measured objects, including three phase-shift patterns for the wrapped phase, and three n-ary gray-level (nGL) patterns for the fringe order. Compared with the binary gray-level (bGL) method which just uses two intensity values, the nGL method can generate many more unique codewords with multiple intensity values. With assistance from the average intensity and modulation of phase-shift patterns, the intensities of nGL patterns are normalized to deal with ambient light and surface contrast. To reduce the codeword detection errors caused by camera/projector defocus, nGL patterns are designed as n-ary gray-code (nGC) patterns to ensure that at most, one code changes at each ...
Dans les domaines de l'apprentissage automatique et de la classification de formes, une tendance ... more Dans les domaines de l'apprentissage automatique et de la classification de formes, une tendance courante est l'exploitation d'énormes quantités d'informations afin de parvenir à une amélioration des performances. En particulier, l'apprentissage à partir d'immenses collections de données collectées sur le Web, ainsi que l'utilisation de plusieurs primitives générées à partir de différentes sources, ont permis d'améliorer de façon significative les résultats de problèmes longtemps considérés comme très difficiles. Dans cette thèse, nous présentons deux façons d'utiliser ces informations pour construire des systèmes d'apprentissage aux performances robustes et avec un certain degré d'autonomie. Ces moyens sont l'intégration de primitives et l'exploitation de primitives, et constituent les deux composantes fondamentales de cette thèse. Dans la première composante, nous présentons plusieurs algorithmes pour répondre à la question portant sur la façon optimale d'intégrer plusieurs primitives. Nous présentons tout d'abord un cadre d'apprentissage en ligne basé sur l'approche d'intégration de haut niveau dans la littérature d'intégration de primitives. Ce cadre peut être mis en oeuvre avec les algorithmes d'apprentissage en ligne existants, et préserve les propriétés théoriques des algorithmes utilisés. Nous avons ensuite étendu le cadre de l'apprentissage par noyaux multiples (MKL), où chaque primitive est convertie en un noyau et où le système apprend un classifieur d'intégration de primitives en résolvant un problème d'optimisation conjointe. Pour rendre le problème réalisable en pratique, nous avons conçu deux nouvelles fonctions de régularisation permettant d'optimiser le problème efficacement. Cela conduisit à la première méthode en ligne pour MKL. Nous montrons également deux algorithmes pour résoudre le problème séquentiel de MKL. Les deux ont une vitesse de convergence garantie. Ces approches atteignent des performances de pointes sur plusieurs jeux de données de référence, et sont un ordre de grandeur plus rapide que les autres trouveurs de solution MKL. Dans la seconde composante, nous présentons deux exemples sur la façon d'exploiter l'information entre différentes sources, afin de réduire l'effort demandé par l'étiquetage d'une grande quantité de données d'entraînement. Le premier exemple est un algorithme d'apprentissage à partir de données partiellement annotées, où chaque donnée est marquée avec plusieurs étiquettes possibles. Nous montrons qu'il est possible de former un système de CONTENTS A Datasets and Features B Miscellaneous Proofs
AbstractObjective. To explore the application results of the clinical departments of public hosp... more AbstractObjective. To explore the application results of the clinical departments of public hospitals based on BSC performance appraisal indicators. Methods. Based on the BSC concept public hospital clinical department indicators of performance appraisal are built and these indicators are used to assess the clinical departments. Results. By comparing the changes before and after the application of relevant indicators, we can see the application of clinical performance assessment indicators does promote the continuous improvement of the quality of medical care. Conclusion. Exploring the finance, customers, internal operational processes, and learning and growth indicators of four dimensions provided guidance for the hospital to develop clinical indicators of performance appraisal ensured a balanced development of hospital quality and efficiency.
2008 IEEE International Conference on Robotics and Automation, 2008
Localization and context interpretation are two key competences for mobile robot systems. Visual ... more Localization and context interpretation are two key competences for mobile robot systems. Visual place recognition, as opposed to purely geometrical models, holds promise of higher flexibility and association of semantics to the model. Ideally, a place recognition algorithm should be robust to dynamic changes and it should perform consistently when recognizing a room (for instance a corridor) in different geographical locations. Also, it should be able to categorize places, a crucial capability for transfer of knowledge and continuous learning. In order to test the suitability of visual recognition algorithms for these tasks, this paper presents a new database, acquired in three different labs across Europe. It contains image sequences of several rooms under dynamic changes, acquired at the same time with a perspective and omnidirectional camera, mounted on a socket. We assess this new database with an appearancebased algorithm that combines local features with support vector machines through an ad-hoc kernel. Results show the effectiveness of the approach and the value of the database.
2007 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2007
Vision-based place recognition is a desirable feature for an autonomous mobile system. In order t... more Vision-based place recognition is a desirable feature for an autonomous mobile system. In order to work in realistic scenarios, visual recognition algorithms should be adaptive, i.e. should be able to learn from experience and adapt continuously to changes in the environment. This paper presents a discriminative incremental learning approach to place recognition. We use a recently introduced version of the incremental SVM, which allows to control the memory requirements as the system updates its internal representation. At the same time, it preserves the recognition performance of the batch algorithm. In order to assess the method, we acquired a database capturing the intrinsic variability of places over time. Extensive experiments show the power and the potential of the approach.
A large amount of images with accompanying text captions are available on the Internet. These are... more A large amount of images with accompanying text captions are available on the Internet. These are valuable for training visual classifiers without any explicit manual intervention. In this paper, we present a general framework to address this problem. Under this new framework, each training image is represented as a bag of regions, associated with a set of candidate labeling vectors. Each labeling vector encodes the possible labels for the regions of the image. The set of all possible labeling vectors can be generated automatically from the caption using natural language processing techniques. The use of labeling vectors provides a principled way to include diverse information from the captions, such as multiple types of words corresponding to different attributes of the same image region, labeling constraints derived from grammatical connections between words, uniqueness constraints, and spatial position indicators. Moreover, it can also be used to incorporate high-level domain knowledge useful for improving learning performance. We show that learning is possible under this weakly supervised setup. Exploiting this property of the problem, we propose a large margin discriminative formulation, and an efficient algorithm to solve the proposed learning problem. Experiments conducted on artificial datasets and two real-world images and captions datasets support our claims.
Proceedings of the 10th international conference on Multimodal interfaces, 2008
It is of prime importance in everyday human life to cope with and respond appropriately to events... more It is of prime importance in everyday human life to cope with and respond appropriately to events that are not foreseen by prior experience. Machines to a large extent lack the ability to respond appropriately to such inputs. An important class of unexpected events is defined by incongruent combinations of inputs from different modalities and therefore multimodal information provides a crucial cue for the identification of such events, e.g., the sound of a voice is being heard while the person in the fieldof-view does not move her lips. In the project DIRAC ("Detection and Identification of Rare Audio-visual Cues") we have been developing algorithmic approaches to the detection of such events, as well as an experimental hardware platform to test it. An audio-visual platform ("AWEAR" -audio-visual wearable device) has been constructed with the goal to help users with disabilities or a high cognitive load to deal with unexpected events. Key hardware components include stereo panoramic vision sensors and 6-channel worn-behind-the-ear (hearing aid) microphone arrays. Data have been recorded to study audio-visual tracking, a/v scene/object classification and a/v detection of incongruencies.
Categorization is one of the fundamental building blocks of cognitive systems. Object categorizat... more Categorization is one of the fundamental building blocks of cognitive systems. Object categorization has traditionally been addressed in the vision domain, even though cognitive agents are intrinsically multimodal. Indeed, biological systems combine several modalities in order to achieve robust categorization. In this paper we propose a multimodal approach to object category detection, using audio and visual information. The auditory channel is modeled on biologically motivated spectral features via a discriminative classifier. The visual channel is modeled by a state of the art part based model. Multimodality is achieved using two fusion schemes, one high level and the other low level. Experiments on six different object categories, under increasingly difficult conditions, show strengths and weaknesses of the two approaches, and clearly underline the open challenges for multimodal category detection.
Procedings of the British Machine Vision Conference 2011, 2011
We present a large scale database of images and captions, designed for supporting research on how... more We present a large scale database of images and captions, designed for supporting research on how to use captioned images from the Web for training visual classifiers. It consists of more than 125,000 images of celebrities from different fields downloaded from the Web. Each image is associated to its original text caption, extracted from the html page the image comes from. We coin it FAN-Large, for Face And Names Large scale database. Its size and deliberate high level of noise makes it to our knowledge the largest and most realistic database supporting this type of research. The dataset and its annotations are publicly available and can be obtained from . We report results on a thorough assessment of FAN-Large using several existing approaches for name-face association, and present and evaluate new contextual features derived from the caption. Our findings provide important cues on the strengths and limitations of existing approaches.
2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2010
Several object categorization algorithms use kernel methods over multiple cues, as they offer a p... more Several object categorization algorithms use kernel methods over multiple cues, as they offer a principled approach to combine multiple cues, and to obtain state-of-theart performance. A general drawback of these strategies is the high computational cost during training, that prevents their application to large-scale problems. They also do not provide theoretical guarantees on their convergence rate. Here we present a Multiclass Multi Kernel Learning (MKL) algorithm that obtains state-of-the-art performance in a considerably lower training time. We generalize the standard MKL formulation to introduce a parameter that allows us to decide the level of sparsity of the solution. Thanks to this new setting, we can directly solve the problem in the primal formulation. We prove theoretically and experimentally that 1) our algorithm has a faster convergence rate as the number of kernels grow; 2) the training complexity is linear in the number of training examples; 3) very few iterations are enough to reach good solutions. Experiments on three standard benchmark databases support our claims.
2011 International Conference on Computer Vision, 2011
The vast majority of transfer learning methods proposed in the visual recognition domain over the... more The vast majority of transfer learning methods proposed in the visual recognition domain over the last years addresses the problem of object category detection, assuming a strong control over the priors from which transfer is done. This is a strict condition, as it concretely limits the use of this type of approach in several settings: for instance, it does not allow in general to use off-the-shelf models as priors. Moreover, the lack of a multiclass formulation for most of the existing transfer learning algorithms prevents using them for object categorization problems, where their use might be beneficial, especially when the number of categories grows and it becomes harder to get enough annotated data for training standard learning methods. This paper presents a multiclass transfer learning algorithm that allows to take advantage of priors built over different features and with different learning methods than the one used for learning the new task. We use the priors as experts, and transfer their outputs to the new incoming samples as additional information. We cast the learning problem within the Multi Kernel Learning framework. The resulting formulation solves efficiently a joint optimization problem that determines from where and how much to transfer, with a principled multiclass formulation. Extensive experiments illustrate the value of this approach.
2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Workshops, 2010
Efficient learning from massive amounts of information is a hot topic in computer vision. Availab... more Efficient learning from massive amounts of information is a hot topic in computer vision. Available training sets contain many examples with several visual descriptors, a setting in which current batch approaches are typically slow and does not scale well. In this work we introduce a theoretically motivated and efficient online learning algorithm for the Multi Kernel Learning (MKL) problem. For this algorithm we prove a theoretical bound on the number of multiclass mistakes made on any arbitrary data sequence. Moreover, we empirically show that its performance is on par, or better, than standard batch MKL (e.g. SILP, Sim-pleMKL) algorithms.
Procedings of the British Machine Vision Conference 2007, 2007
In the framework of indoor mobile robotics, place recognition is a challenging task, where it is ... more In the framework of indoor mobile robotics, place recognition is a challenging task, where it is crucial that self-localization be enforced precisely, notwithstanding the changing conditions of illumination, objects being shifted around and/or people affecting the appearance of the scene. In this scenario online learning seems the main way out, thanks to the possibility of adapting to changes in a smart and flexible way. Nevertheless, standard machine learning approaches usually suffer when confronted with massive amounts of data and when asked to work online. Online learning requires a high training and testing speed, all the more in place recognition, where a continuous flow of data comes from one or more cameras. In this paper we follow the Support Vector Machines-based approach of Pronobis et al. , proposing an improvement that we call Online Independent Support Vector Machines. This technique exploits linear independence in the image feature space to incrementally keep the size of the learning machine remarkably small while retaining the accuracy of a standard machine. Since the training and testing time crucially depend on the size of the machine, this solves the above stated problems. Our experimental results prove the effectiveness of the approach.
Support Vector Machines (SVMs) are one of the most successful algorithms for classification. Howe... more Support Vector Machines (SVMs) are one of the most successful algorithms for classification. However, due to their space and time requirements, they are not suitable for on-line learning, that is, when presented with an endless stream of training observations. In this paper we propose a new on-line algorithm, called On-line Independent Support Vector Machines (OISVMs), which approximately converges to the standard SVM solution each time new observations are added; the approximation is controlled via a user-defined parameter. The method employs a set of linearly independent observations and tries to project every new observation onto the set obtained so far, dramatically reducing time and space requirements at the price of a negligible loss in accuracy. As opposed to similar algorithms, the size of the solution obtained by OISVMs is always bounded, implying a bounded testing time. These statements are supported by extensive experiments on standard benchmark databases as well as on two real-world applications, namely place recognition
By using one-bit binary patterns instead of eight-bit sinusoidal ones, the binary defocusing tech... more By using one-bit binary patterns instead of eight-bit sinusoidal ones, the binary defocusing techniques have been widely applied for high-speed 3D shape measurement. As projector defocusing is required, the phase unwrapping process of these techniques remains challenging. A recently proposed ternary Gray-code method can effectively increase the measurement speed by reducing the number of acquired coding patterns. However, it still has limitations, including the measuring range and a noise problem. To improve these, a new ternary Gray-code method is proposed by utilizing 2D modulation instead of the 1D modulation used in the conventional method. Simulations were conducted to compare these two methods, and the results verify that the proposed method reduces the variance in the introduced intermediate gray value by nearly 90%. Experiments were also carried out with the results verifying the superiority of the proposed method.
The vehicle dynamics model has multiple degrees of freedom, with strong nonlinear characteristics... more The vehicle dynamics model has multiple degrees of freedom, with strong nonlinear characteristics, so it is difficult to quickly obtain the accurate target oil pressure of an electronically assisted brake system based on the model. This paper proposes a target oil pressure recognition algorithm based on the T-S fuzzy neural network model. Firstly, the braking conditions classification algorithm is built according to the sampled braking intention data. The data are divided into the emergency braking condition data and the general braking condition data by the braking conditions classification algorithm. Secondly, the recognition model is trained respectively by the different braking condition data sets. In the training process, the fuzzy C-means clustering algorithm is used to identify the antecedent parameters of the model, and the learning rate cosine attenuation strategy is applied to optimize the parameter learning process. Finally, a correction method of target oil pressure base...
This paper presents an SVM--based algorithm for the transfer of knowledge across robot platforms ... more This paper presents an SVM--based algorithm for the transfer of knowledge across robot platforms aiming to perform the same task. Our method exploits efficiently the transferred knowledge while updating incrementally the internal representation as new information is available. The algorithm is adaptive and tends to privilege new data when building the SV solution. This prevents the old knowledge to nest into the model and eventually become a possible source of misleading information. We tested our approach in the domain of vision-based place recognition. Extensive experiments show that using transferred knowledge clearly pays off in terms of performance and stability of the solution.
ELCVIA Electronic Letters on Computer Vision and Image Analysis, 2010
Local features have repeatedly shown their effectiveness for object recognition during the last y... more Local features have repeatedly shown their effectiveness for object recognition during the last years, and they have consequently become the preferred descriptor for this type of problems. The solution of the correspondence problem is traditionally approached with exact or approximate techniques. In this paper we are interested in methods that solve the correspondence problem via the definition of a kernel function that makes it possible to use local features as input to a support vector machine. We single out the match kernel, an exact approach, and the pyramid match kernel, that uses instead an approximate strategy. We present a thorough experimental evaluation of the two methods on three different databases. Results show that the exact method performs consistently better than the approximate one, especially for the object identification task, when training on a decreasing number of images. Based on these findings and on the computational cost of each approach, we suggest some criteria for choosing between the two kernels given the application at hand.
The accurate air-fuel ratio (AFR) control is crucial for the exhaust emission reduction based on ... more The accurate air-fuel ratio (AFR) control is crucial for the exhaust emission reduction based on the three-way catalytic converter in the spark ignition (SI) engine. The difficulties in transient cylinder air mass flow measurement, the existing fuel mass wall-wetting phenomenon, and the unfixed AFR path dynamic variations make the design of the AFR controller a challenging task. In this paper, an adaptive AFR regulation controller is designed using the feedforward and feedback control scheme based on the dynamical modelling of the AFR path. The generalized predictive control method is proposed to solve the problems of inherent nonlinearities, time delays, parameter variations, and uncertainties in the AFR closed loop. The simulation analysis is investigated for the effectiveness of noise suppression, online prediction, and self-correction on the SI engine system. Moreover, the experimental verification shows an acceptable performance of the designed controller and the potential usag...
Fringe projection technologies have been widely used for three-dimensional (3D) shape measurement... more Fringe projection technologies have been widely used for three-dimensional (3D) shape measurement. One of the critical issues is absolute phase recovery, especially for measuring multiple isolated objects. This paper proposes a method for absolute phase retrieval using only one coded pattern. A total of four patterns including one coded pattern and three phase-shift patterns are projected, captured, and processed. The wrapped phase, as well as average intensity and intensity modulation, are calculated from three phase-shift patterns. A code word encrypted into the coded pattern can be calculated using the average intensity and intensity modulation. Based on geometric constraints of fringe projection system, the minimum fringe order map can be created, upon which the fringe order can be calculated from the code word. Compared with the conventional method, the measurement depth range is significantly improved. Finally, the wrapped phase can be unwrapped for absolute phase map. Since o...
Fringe projection systems have been widely applied in three-dimensional (3D) shape measurements. ... more Fringe projection systems have been widely applied in three-dimensional (3D) shape measurements. One of the important issues is how to retrieve the absolute phase. This paper presents a modified gray-level coding method for absolute phase retrieval. Specifically, two groups of fringe patterns are projected onto the measured objects, including three phase-shift patterns for the wrapped phase, and three n-ary gray-level (nGL) patterns for the fringe order. Compared with the binary gray-level (bGL) method which just uses two intensity values, the nGL method can generate many more unique codewords with multiple intensity values. With assistance from the average intensity and modulation of phase-shift patterns, the intensities of nGL patterns are normalized to deal with ambient light and surface contrast. To reduce the codeword detection errors caused by camera/projector defocus, nGL patterns are designed as n-ary gray-code (nGC) patterns to ensure that at most, one code changes at each ...
Dans les domaines de l'apprentissage automatique et de la classification de formes, une tendance ... more Dans les domaines de l'apprentissage automatique et de la classification de formes, une tendance courante est l'exploitation d'énormes quantités d'informations afin de parvenir à une amélioration des performances. En particulier, l'apprentissage à partir d'immenses collections de données collectées sur le Web, ainsi que l'utilisation de plusieurs primitives générées à partir de différentes sources, ont permis d'améliorer de façon significative les résultats de problèmes longtemps considérés comme très difficiles. Dans cette thèse, nous présentons deux façons d'utiliser ces informations pour construire des systèmes d'apprentissage aux performances robustes et avec un certain degré d'autonomie. Ces moyens sont l'intégration de primitives et l'exploitation de primitives, et constituent les deux composantes fondamentales de cette thèse. Dans la première composante, nous présentons plusieurs algorithmes pour répondre à la question portant sur la façon optimale d'intégrer plusieurs primitives. Nous présentons tout d'abord un cadre d'apprentissage en ligne basé sur l'approche d'intégration de haut niveau dans la littérature d'intégration de primitives. Ce cadre peut être mis en oeuvre avec les algorithmes d'apprentissage en ligne existants, et préserve les propriétés théoriques des algorithmes utilisés. Nous avons ensuite étendu le cadre de l'apprentissage par noyaux multiples (MKL), où chaque primitive est convertie en un noyau et où le système apprend un classifieur d'intégration de primitives en résolvant un problème d'optimisation conjointe. Pour rendre le problème réalisable en pratique, nous avons conçu deux nouvelles fonctions de régularisation permettant d'optimiser le problème efficacement. Cela conduisit à la première méthode en ligne pour MKL. Nous montrons également deux algorithmes pour résoudre le problème séquentiel de MKL. Les deux ont une vitesse de convergence garantie. Ces approches atteignent des performances de pointes sur plusieurs jeux de données de référence, et sont un ordre de grandeur plus rapide que les autres trouveurs de solution MKL. Dans la seconde composante, nous présentons deux exemples sur la façon d'exploiter l'information entre différentes sources, afin de réduire l'effort demandé par l'étiquetage d'une grande quantité de données d'entraînement. Le premier exemple est un algorithme d'apprentissage à partir de données partiellement annotées, où chaque donnée est marquée avec plusieurs étiquettes possibles. Nous montrons qu'il est possible de former un système de CONTENTS A Datasets and Features B Miscellaneous Proofs
AbstractObjective. To explore the application results of the clinical departments of public hosp... more AbstractObjective. To explore the application results of the clinical departments of public hospitals based on BSC performance appraisal indicators. Methods. Based on the BSC concept public hospital clinical department indicators of performance appraisal are built and these indicators are used to assess the clinical departments. Results. By comparing the changes before and after the application of relevant indicators, we can see the application of clinical performance assessment indicators does promote the continuous improvement of the quality of medical care. Conclusion. Exploring the finance, customers, internal operational processes, and learning and growth indicators of four dimensions provided guidance for the hospital to develop clinical indicators of performance appraisal ensured a balanced development of hospital quality and efficiency.
2008 IEEE International Conference on Robotics and Automation, 2008
Localization and context interpretation are two key competences for mobile robot systems. Visual ... more Localization and context interpretation are two key competences for mobile robot systems. Visual place recognition, as opposed to purely geometrical models, holds promise of higher flexibility and association of semantics to the model. Ideally, a place recognition algorithm should be robust to dynamic changes and it should perform consistently when recognizing a room (for instance a corridor) in different geographical locations. Also, it should be able to categorize places, a crucial capability for transfer of knowledge and continuous learning. In order to test the suitability of visual recognition algorithms for these tasks, this paper presents a new database, acquired in three different labs across Europe. It contains image sequences of several rooms under dynamic changes, acquired at the same time with a perspective and omnidirectional camera, mounted on a socket. We assess this new database with an appearancebased algorithm that combines local features with support vector machines through an ad-hoc kernel. Results show the effectiveness of the approach and the value of the database.
2007 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2007
Vision-based place recognition is a desirable feature for an autonomous mobile system. In order t... more Vision-based place recognition is a desirable feature for an autonomous mobile system. In order to work in realistic scenarios, visual recognition algorithms should be adaptive, i.e. should be able to learn from experience and adapt continuously to changes in the environment. This paper presents a discriminative incremental learning approach to place recognition. We use a recently introduced version of the incremental SVM, which allows to control the memory requirements as the system updates its internal representation. At the same time, it preserves the recognition performance of the batch algorithm. In order to assess the method, we acquired a database capturing the intrinsic variability of places over time. Extensive experiments show the power and the potential of the approach.
A large amount of images with accompanying text captions are available on the Internet. These are... more A large amount of images with accompanying text captions are available on the Internet. These are valuable for training visual classifiers without any explicit manual intervention. In this paper, we present a general framework to address this problem. Under this new framework, each training image is represented as a bag of regions, associated with a set of candidate labeling vectors. Each labeling vector encodes the possible labels for the regions of the image. The set of all possible labeling vectors can be generated automatically from the caption using natural language processing techniques. The use of labeling vectors provides a principled way to include diverse information from the captions, such as multiple types of words corresponding to different attributes of the same image region, labeling constraints derived from grammatical connections between words, uniqueness constraints, and spatial position indicators. Moreover, it can also be used to incorporate high-level domain knowledge useful for improving learning performance. We show that learning is possible under this weakly supervised setup. Exploiting this property of the problem, we propose a large margin discriminative formulation, and an efficient algorithm to solve the proposed learning problem. Experiments conducted on artificial datasets and two real-world images and captions datasets support our claims.
Proceedings of the 10th international conference on Multimodal interfaces, 2008
It is of prime importance in everyday human life to cope with and respond appropriately to events... more It is of prime importance in everyday human life to cope with and respond appropriately to events that are not foreseen by prior experience. Machines to a large extent lack the ability to respond appropriately to such inputs. An important class of unexpected events is defined by incongruent combinations of inputs from different modalities and therefore multimodal information provides a crucial cue for the identification of such events, e.g., the sound of a voice is being heard while the person in the fieldof-view does not move her lips. In the project DIRAC ("Detection and Identification of Rare Audio-visual Cues") we have been developing algorithmic approaches to the detection of such events, as well as an experimental hardware platform to test it. An audio-visual platform ("AWEAR" -audio-visual wearable device) has been constructed with the goal to help users with disabilities or a high cognitive load to deal with unexpected events. Key hardware components include stereo panoramic vision sensors and 6-channel worn-behind-the-ear (hearing aid) microphone arrays. Data have been recorded to study audio-visual tracking, a/v scene/object classification and a/v detection of incongruencies.
Categorization is one of the fundamental building blocks of cognitive systems. Object categorizat... more Categorization is one of the fundamental building blocks of cognitive systems. Object categorization has traditionally been addressed in the vision domain, even though cognitive agents are intrinsically multimodal. Indeed, biological systems combine several modalities in order to achieve robust categorization. In this paper we propose a multimodal approach to object category detection, using audio and visual information. The auditory channel is modeled on biologically motivated spectral features via a discriminative classifier. The visual channel is modeled by a state of the art part based model. Multimodality is achieved using two fusion schemes, one high level and the other low level. Experiments on six different object categories, under increasingly difficult conditions, show strengths and weaknesses of the two approaches, and clearly underline the open challenges for multimodal category detection.
Procedings of the British Machine Vision Conference 2011, 2011
We present a large scale database of images and captions, designed for supporting research on how... more We present a large scale database of images and captions, designed for supporting research on how to use captioned images from the Web for training visual classifiers. It consists of more than 125,000 images of celebrities from different fields downloaded from the Web. Each image is associated to its original text caption, extracted from the html page the image comes from. We coin it FAN-Large, for Face And Names Large scale database. Its size and deliberate high level of noise makes it to our knowledge the largest and most realistic database supporting this type of research. The dataset and its annotations are publicly available and can be obtained from . We report results on a thorough assessment of FAN-Large using several existing approaches for name-face association, and present and evaluate new contextual features derived from the caption. Our findings provide important cues on the strengths and limitations of existing approaches.
2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2010
Several object categorization algorithms use kernel methods over multiple cues, as they offer a p... more Several object categorization algorithms use kernel methods over multiple cues, as they offer a principled approach to combine multiple cues, and to obtain state-of-theart performance. A general drawback of these strategies is the high computational cost during training, that prevents their application to large-scale problems. They also do not provide theoretical guarantees on their convergence rate. Here we present a Multiclass Multi Kernel Learning (MKL) algorithm that obtains state-of-the-art performance in a considerably lower training time. We generalize the standard MKL formulation to introduce a parameter that allows us to decide the level of sparsity of the solution. Thanks to this new setting, we can directly solve the problem in the primal formulation. We prove theoretically and experimentally that 1) our algorithm has a faster convergence rate as the number of kernels grow; 2) the training complexity is linear in the number of training examples; 3) very few iterations are enough to reach good solutions. Experiments on three standard benchmark databases support our claims.
2011 International Conference on Computer Vision, 2011
The vast majority of transfer learning methods proposed in the visual recognition domain over the... more The vast majority of transfer learning methods proposed in the visual recognition domain over the last years addresses the problem of object category detection, assuming a strong control over the priors from which transfer is done. This is a strict condition, as it concretely limits the use of this type of approach in several settings: for instance, it does not allow in general to use off-the-shelf models as priors. Moreover, the lack of a multiclass formulation for most of the existing transfer learning algorithms prevents using them for object categorization problems, where their use might be beneficial, especially when the number of categories grows and it becomes harder to get enough annotated data for training standard learning methods. This paper presents a multiclass transfer learning algorithm that allows to take advantage of priors built over different features and with different learning methods than the one used for learning the new task. We use the priors as experts, and transfer their outputs to the new incoming samples as additional information. We cast the learning problem within the Multi Kernel Learning framework. The resulting formulation solves efficiently a joint optimization problem that determines from where and how much to transfer, with a principled multiclass formulation. Extensive experiments illustrate the value of this approach.
2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Workshops, 2010
Efficient learning from massive amounts of information is a hot topic in computer vision. Availab... more Efficient learning from massive amounts of information is a hot topic in computer vision. Available training sets contain many examples with several visual descriptors, a setting in which current batch approaches are typically slow and does not scale well. In this work we introduce a theoretically motivated and efficient online learning algorithm for the Multi Kernel Learning (MKL) problem. For this algorithm we prove a theoretical bound on the number of multiclass mistakes made on any arbitrary data sequence. Moreover, we empirically show that its performance is on par, or better, than standard batch MKL (e.g. SILP, Sim-pleMKL) algorithms.
Procedings of the British Machine Vision Conference 2007, 2007
In the framework of indoor mobile robotics, place recognition is a challenging task, where it is ... more In the framework of indoor mobile robotics, place recognition is a challenging task, where it is crucial that self-localization be enforced precisely, notwithstanding the changing conditions of illumination, objects being shifted around and/or people affecting the appearance of the scene. In this scenario online learning seems the main way out, thanks to the possibility of adapting to changes in a smart and flexible way. Nevertheless, standard machine learning approaches usually suffer when confronted with massive amounts of data and when asked to work online. Online learning requires a high training and testing speed, all the more in place recognition, where a continuous flow of data comes from one or more cameras. In this paper we follow the Support Vector Machines-based approach of Pronobis et al. , proposing an improvement that we call Online Independent Support Vector Machines. This technique exploits linear independence in the image feature space to incrementally keep the size of the learning machine remarkably small while retaining the accuracy of a standard machine. Since the training and testing time crucially depend on the size of the machine, this solves the above stated problems. Our experimental results prove the effectiveness of the approach.
Support Vector Machines (SVMs) are one of the most successful algorithms for classification. Howe... more Support Vector Machines (SVMs) are one of the most successful algorithms for classification. However, due to their space and time requirements, they are not suitable for on-line learning, that is, when presented with an endless stream of training observations. In this paper we propose a new on-line algorithm, called On-line Independent Support Vector Machines (OISVMs), which approximately converges to the standard SVM solution each time new observations are added; the approximation is controlled via a user-defined parameter. The method employs a set of linearly independent observations and tries to project every new observation onto the set obtained so far, dramatically reducing time and space requirements at the price of a negligible loss in accuracy. As opposed to similar algorithms, the size of the solution obtained by OISVMs is always bounded, implying a bounded testing time. These statements are supported by extensive experiments on standard benchmark databases as well as on two real-world applications, namely place recognition
Uploads
Papers by LUO JIE