Abstract: A search apparatus for searching media data for a target region that matches query data includes a first feature extraction unit configured to extract a first feature vector from the query data using a first trained neural network; a second feature extraction unit configured to obtain a first region from the media data and extract a second feature vector from the first region using a second trained neural network; a localization unit configured to determine a candidate for the target region using a third trained neural network, based on the first feature vector, the second feature vector, and the first region or a location of the first region; and a control unit configured to repeat the operations of the second feature extraction unit and the localization unit until a predetermined condition is satisfied, by using the determined candidate for the target region as the first region to be used by the second feature extraction unit.

Abstract: An acquiring unit of a depth estimation apparatus acquires an input image. In addition, a depth map generating unit inputs the input image acquired by the acquiring unit into a depth estimator for generating, from an image, a depth map in which a depth of a space that appears on the image is imparted to each pixel of the image, and generates an estimated depth map that represents a depth map corresponding to the input image. The depth estimator is a model having been learned in advance so as to reduce, with respect to each error between a depth of the estimated depth map and a depth of a correct-answer depth map that presents the depth map of a correct answer, a value of a loss function set such that a degree of increase of a loss value with respect to a pixel at which the error is larger than a threshold is smaller than a degree of increase of a loss value with respect to a pixel at which the error is equal to or smaller than the threshold.

Abstract: A learning device includes: a data input unit configured to receive first data which is a learning target, second data for identifying the first data, and second past data that is used as data for identifying the first data during past learning and relates to learning content to be preserved; a combined data generation unit configured to generate combined data by combining the first data and the second data; and a parameter updating unit configured to update a parameter of a machine learning model based on features of the second past data and the combined data obtained by inputting the combined data and the second past data to the machine learning model.

Abstract: An image identification device can be trained to identify classes with high accuracy even in cases with a small number of learning images.

Abstract: The purpose of the present invention is to enable learning of a neural network for extracting features of images having high robustness from an undiscriminating image region while minimizing the number of parameters of a pooling layer. A parameter learning unit 130 learns parameters of each layer in a convolutional neural network configured by including a fully convolutional layer for performing convolution of an input image to output a feature tensor of the input image, a weighting matrix estimation layer for estimating a weighting matrix indicating a weighting of each element of the feature tensor, and a pooling layer for extracting a feature vector of the input image based on the feature tensor and the weighting matrix.

Abstract: It is possible to receive a point cloud as input and perform shape completion with high accuracy. A shape completion unit inputs an input point cloud and a class identification feature output by a class identification unit to a generator that is learned in advance and generates a shape completion point cloud that is to complete a point cloud and is a set of three-dimensional points by receiving, as input, the point cloud and the class identification feature, gaining an integration result obtained by integrating a global feature that is a global feature based on local features extracted from respective points of the point cloud with the class identification feature, and convoluting the integration result, and outputs the shape completion point cloud that is to complete the input point cloud.

Abstract: A depth estimation method using a depth estimator trained to output a depth map of a depth provided to each pixel of an input image, in which: the depth estimator includes a pair of a first convolutional layer and a second convolutional layer coupled to each other and configured to, when having received, as input, a tensor obtained by applying predetermined conversion to an input image, apply a two-dimensional convolution operation to the tensor and output the tensor to which the two-dimensional convolution operation is applied; the first convolutional layer is a convolutional layer including a first kernel of a shape having lengths in a first direction and a second direction, the first direction being one of a vertical direction and a horizontal direction, the second direction being different from the first direction, the length in the second direction being longer than the length in the first direction; and the second convolutional layer is a convolutional layer including a second kernel of a shape having l

Abstract: Low-dimensional feature values with which semantic factors of content are ascertained are generated from relevance between sets of two types of content. Based on a relation indicator indicating a pair of groups indicating which groups are related to first types of content groups among second types of content groups, an initial feature value extracting unit 11 extracts initial feature values of the first type of content and the second type of content. A content pair selecting unit 12 selects a content pair by selecting one first type of content and one second type of content from each pair of groups indicated by the relation indicator. A feature value conversion function generating unit 13 generates feature conversion functions 31 of converting the initial feature values into low-dimensional feature values based on the content pair selected from each pair of groups.

Abstract: Clustering can be performed using a self-expression matrix in which noise is suppressed. A self-expression matrix is calculated that minimizes an objective function that is for obtaining, from among matrices included in a predetermined matrix set, a self-expression matrix whose elements are linear weights when data points in a data set are expressed by linear combinations of points, the objective function being represented by a term for obtaining the residual between data points in the data set and data points expressed by linear combinations of points using the self-expression matrix, a first regularization term that is multiplied by a predetermined weight and is for reducing linear weights of the data points that have a large Euclidean norm in the self-expression matrix, and a second regularization term for the self-expression matrix. A similarity matrix defined by the calculated self-expression matrix is then calculated.

Abstract: In a depth estimation device, a generation unit generates a predetermined attractive sound in a space to be measured. A sound pickup unit picks up an acoustic signal for a predetermined time period corresponding to a time period before and after a time of generation of the attractive sound. An estimation unit extracts a feature representing time-frequency information obtained through analysis of the acoustic signal, on the basis of the acoustic signal, and inputs the extracted feature representing the time-frequency information to a depth estimator and generates an estimated depth map for the space to be measured, the depth estimator being composed of one or more convolution operations and being learned so as to output an estimated depth map, in which a depth is assigned to each of pixels of an image representing the space to be measured, when a feature representing the time-frequency information is input.

Abstract: An acquiring unit of a depth estimation apparatus acquires an input image. In addition, a depth map generating unit inputs the input image acquired by the acquiring unit into a depth estimator for generating, from an image, a depth map in which a depth of a space that appears on the image is imparted to each pixel of the image, and generates an estimated depth map that represents a depth map corresponding to the input image. The depth estimator is a model having been learned in advance so as to reduce, with respect to each error between a depth of the estimated depth map and a depth of a correct-answer depth map that presents the depth map of a correct answer, a value of a loss function set such that a degree of increase of a loss value with respect to a pixel at which the error is larger than a threshold is smaller than a degree of increase of a loss value with respect to a pixel at which the error is equal to or smaller than the threshold.

Abstract: A search apparatus for searching media data for a target region that matches query data includes a first feature extraction unit configured to extract a first feature vector from the query data using a first trained neural network; a second feature extraction unit configured to obtain a first region from the media data and extract a second feature vector from the first region using a second trained neural network; a localization unit configured to determine a candidate for the target region using a third trained neural network, based on the first feature vector, the second feature vector, and the first region or a location of the first region; and a control unit configured to repeat the operations of the second feature extraction unit and the localization unit until a predetermined condition is satisfied, by using the determined candidate for the target region as the first region to be used by the second feature extraction unit.

Abstract: It is possible to receive a point cloud as input and perform shape completion with high accuracy. A shape completion unit inputs an input point cloud and a class identification feature output by a class identification unit to a generator that is learned in advance and generates a shape completion point cloud that is to complete a point cloud and is a set of three-dimensional points by receiving, as input, the point cloud and the class identification feature, gaining an integration result obtained by integrating a global feature that is a global feature based on local features extracted from respective points of the point cloud with the class identification feature, and convoluting the integration result, and outputs the shape completion point cloud that is to complete the input point cloud.

Abstract: Clustering can be performed using a self-expression matrix in which noise is suppressed. A self-expression matrix is calculated that minimizes an objective function that is for obtaining, from among matrices included in a predetermined matrix set, a self-expression matrix whose elements are linear weights when data points in a data set are expressed by linear combinations of points, the objective function being represented by a term for obtaining the residual between data points in the data set and data points expressed by linear combinations of points using the self-expression matrix, a first regularization term that is multiplied by a predetermined weight and is for reducing linear weights of the data points that have a large Euclidean norm in the self-expression matrix, and a second regularization term for the self-expression matrix. A similarity matrix defined by the calculated self-expression matrix is then calculated.

Abstract: An image identification device can be trained to identify classes with high accuracy even in cases with a small number of learning images.

Abstract: Low-dimensional feature values with which semantic factors of content are ascertained are generated from relevance between sets of two types of content. Based on a relation indicator indicating a pair of groups indicating which groups are related to first types of content groups among second types of content groups, an initial feature value extracting unit 11 extracts initial feature values of the first type of content and the second type of content. A content pair selecting unit 12 selects a content pair by selecting one first type of content and one second type of content from each pair of groups indicated by the relation indicator. A feature value conversion function generating unit 13 generates feature conversion functions 31 of converting the initial feature values into low-dimensional feature values based on the content pair selected from each pair of groups.

Abstract: The purpose of the present invention is to enable learning of a neural network for extracting features of images having high robustness from an undiscriminating image region while minimizing the number of parameters of a pooling layer. A parameter learning unit 130 learns parameters of each layer in a convolutional neural network configured by including a fully convolutional layer for performing convolution of an input image to output a feature tensor of the input image, a weighting matrix estimation layer for estimating a weighting matrix indicating a weighting of each element of the feature tensor, and a pooling layer for extracting a feature vector of the input image based on the feature tensor and the weighting matrix.

Abstract: An audio feature is extracted from audio signal data for each analysis frame and stored in a storage part. Then, the audio feature is read from the storage part, and an emotional state probability of the audio feature corresponding to an emotional state is calculated using one or more statistical models constructed based on previously input learning audio signal data. Then, based on the calculated emotional state probability, the emotional state of a section including the analysis frame is determined.

Abstract: An audio feature is extracted from audio signal data for each analysis frame and stored in a storage part. Then, the audio feature is read from the storage part, and an emotional state probability of the audio feature corresponding to an emotional state is calculated using one or more statistical models constructed based on previously input learning audio signal data. Then, based on the calculated emotional state probability, the emotional state of a section including the analysis frame is determined.