Fujitsu Corporation

Rujie Liu

Yusuke Uehara

Meng Zhang

Takayuki Baba

Susumu Endo

Daiki Masumoto

Wei Shen

Shuichi Shiitani

Shigemi Nagata

Mengjiao Wang

Qiong Cao

Hao Yu

Liu Liu

Fei Fei Li

Pingping Lu

Ziqiang Shi

Shuchi Shiitani

Jun Sun

Masahiko Sugimura

Xi Feng Liu

Yuehong Wang

Dai Miyawaki

Song Guo

A method and an apparatus for training a neural network, an image recognition method and a computer readable storage medium are disclosed. The neural network includes a first model and a second model. The method for training a neural network includes: acquiring a second image from a first image, wherein a quality of the second image is lower than that of the first image; inputting the first image into the first model of the neural network, and inputting the second image into the second model of the neural network; calculating an attention map and a gradient map of the first model and an attention map and a gradient map of the second model; constructing a loss function based on a matrix of a dot product of the gradient map and the attention map of the first model and a matrix of a dot product of the gradient map and the attention map of the second model; and training the neural network by minimizing the loss function.

Method and apparatus for training a neural network, image recognition method and storage medium

A device and a method for classification using a pre-trained classification model and a computer readable storage medium are provided. The device is configured to extract, for each of multiple images in a target image group to be classified, a feature of the image using a feature extraction layer of the pre-trained classification model; calculate, for each of the multiple images, a contribution of the image to a classification result of the target image group using a contribution calculation layer of the pre-trained classification model; aggregate extracted features of the multiple images based on calculated contributions of the multiple images, to obtain an aggregated feature as a feature of the target image group; and classify the target image group based on the feature of the target image group.

Device and method for classification using classification model and computer readable storage medium

There are provided a method of knowledge transferring, an information processing apparatus and a storage medium. The method of knowledge transferring includes: obtaining a first model which has been trained in advance with respect to a predetermined task; and training a second model with respect to the predetermined task by utilizing a comprehensive loss function, such that the second model has knowledge of the first model. The comprehensive loss function is based on a first loss function weighted by accuracy of an output result of the first model for a training sample in regard to the predetermined task, and a second loss function. The first loss function represents a difference between processing results of the second model and the first model for the training sample. The second loss function represents accuracy of an output result of the second model for the training sample in regard to the predetermined task.

Method of knowledge transferring, information processing apparatus and storage medium

An apparatus for training a classification model includes: a feature extraction unit configured to set, with respect to each training set of a first predetermined number of training sets, feature extraction layers, and extract features of a sample image, where at least two of the training sets at least partially overlap; a feature fusion unit configured to set, with respect to training set, feature fusion layers, and perform a fusion on the extracted features of the sample image; and a loss determination unit configured to set, with respect to training set, a loss determination layer, calculate a loss function of the sample image based on the fused feature of the sample image, and train a classification model based on the loss function. The first predetermined number of training sets share at least one layer of feature fusion layers and feature extraction layers set with respect to each training set.

Apparatus and method for training classification model and apparatus for performing classification by using classification model

A method for training a classification model includes: performing training on the classification model using first and second sample sets, to calculate a classification loss; extracting a weight vector and a feature vector of each sample; calculating a mean weight vector and a mean feature vector of all samples in the first sample set; calculating a weight loss based on a difference of the weight vector of each sample in the second sample set from the mean weight vector, and calculating a feature loss based on a difference of a feature vector of each sample in the second sample set from the mean feature vector; calculating a total loss of the classification model based on the classification loss and at least one of the feature loss and the weight loss; and adjusting a parameter of the classification model until a predetermined condition is satisfied.

Method and apparatus for training classification model, and classification method

An information processing apparatus includes a processor to input each sample image into feature extracting components to obtain at least two features of the sample image, and to cause a classifying component to calculate a classification loss of the sample image based on the at least two features; extract, from each pair of features, a plurality of sample pairs for calculating mutual information between each pair of features; input the plurality of sample pairs into a machine learning architecture corresponding to each pair of features, to calculate an information loss between each pair of features. The processor is to adjust parameters of the feature extracting components, the classifying component and the machine learning architecture by minimizing a sum of classification losses and information losses of sample images in the training set based upon the obtained at least two features of the sample image to calculate the classification losses and the information losses, to obtain the at least two feature extracting components and the classifying component having been trained.

Information processing apparatus and information processing method

An apparatus for training a classifying model comprises: a first obtaining unit configured to input a sample image to a first machine learning framework, to obtain a first classification probability and a first classification loss; a second obtaining unit configured to input a second image to a second machine learning framework, to obtain a second classification probability and a second classification loss, the two machine learning frameworks having identical structures and sharing identical parameters; a similarity loss calculating unit configured to calculate a similarity loss related to a similarity between the first classification probability and the second classification probability; a total loss calculating unit configured to calculate the sum of the similarity loss, the first classification loss and the second classification loss, as a total loss; and a training unit configured to adjust parameters of the two machine learning frameworks to obtain a trained classifying model.

Apparatus and method for training classifying model

An information processing method includes: inputting sample image into a machine learning architecture to obtain a first feature, and causing a first classifier to calculate a first classification loss; calculating a second feature based on the first feature and a predetermined first mask, and inputting the second feature into the first classifier to calculate an entropy loss; calculating a second mask based on the first mask and the entropy loss to maximize the entropy loss; obtaining an adversarial feature based on the first feature and the second mask, where the adversarial feature is complementary to the second feature; causing, by training the first classifier and the second classifier in association with each other, the second classifier to calculate a second classification loss based on the adversarial feature; and adjusting parameters of the machine learning architecture, the first classifier and the second classifier, to obtain a trained machine learning architecture.

An apparatus and method for training a classification model and an apparatus for classifying with a classification model are disclosed. The apparatus for training a classification model comprises: a local area obtainment unit to, obtain predetermined local area of each sample image; a feature extraction unit to, with respect to each sample image, set corresponding numbers of feature extraction layers for the global area and each predetermined local area, to extract a global feature of the global area and a local feature of each predetermined local area, wherein the global area and the predetermined local areas share at least one feature extraction layer in which the global feature and each local feature are combined; and a loss determination unit to calculate a loss function of the sample image based on combined features of each sample image, and to train the classification model based on the loss function.

Apparatus and method for training classification model and apparatus for classifying with classification model

The present disclosure relates to an information processing method and an information processing apparatus. The information processing method according to the present disclosure performs training on a classification model by using a plurality of training samples, and comprises the steps of: adjusting a distribution of feature vectors of the plurality of training samples in a feature space based on a typical sample in the plurality of training samples; and performing training on the classification model by using the adjusted feature vectors of the plurality of training samples. Through the technology according to the present disclosure, it is possible to perform pre-adjustment on training samples before training, such that it is possible to reduce discrimination between training samples belonging to a same class and increase discrimination between training samples belonging to different classes in the training process. The classification model trained as such is capable of performing accurate classification on samples acquired under an extreme condition.

Information processing method and information processing apparatus

An information processing method and an information processing apparatus are disclosed, where the information processing method includes: inputting a plurality of samples to a classifier respectively, to extract a feature vector representing a feature of each sample; and updating parameters of the classifier by minimizing a loss function for the plurality of samples, wherein the loss function is in positive correlation with an intra-class distance for representing a distance between feature vectors of samples belonging to a same class, and is in negative correlation with an inter-class distance for representing a distance between feature vectors of samples belonging to different classes, wherein the intra-class distance of each sample of the plurality of samples is less than a first threshold, the inter-class distance between two different classes is greater than a second threshold, and the second threshold is greater than twice the first threshold.

Information processing method and information processing apparatus for improving the discriminality of features of extracted samples

An apparatus and a method for data processing are provided. The apparatus for data processing includes a modeler configured to build an occlusion object model for an image containing an occlusion object; a renderer configured to render the occlusion object model according to a geometric relationship between the occlusion object and a face image containing no occlusion object, such that the rendered occlusion object image and the face image containing no occlusion object have same scale and attitude; and a merger configured to merge the face image containing no occlusion object and the rendered occlusion object image into an occluded face image. With the data processing apparatus and the data processing method face data enhancement, face data in the case of having an occlusion object is generated, so that the number of face training data sets can be effectively increased, thereby improving performance of a face-related module.

Apparatus and method for data processing

A method and apparatus for evaluating an illumination condition in a face image is provided by decomposing a face image into illumination feature components and face feature components; extracting determined areas in the face image; calculating a maximum luminance feature, a minimum luminance feature and an illumination direction feature according to the decomposed illumination feature components in the determined areas. The illumination condition in the face image is evaluated according to the maximum luminance feature, the minimum luminance feature and the illumination direction feature.

Method and apparatus for evaluating illumination condition in face image

The method for training an image model, in each round of training performed with respect to each sample image: inputs an image obtained by cropping the sample image by an object extraction component obtained through a previous round of training, as a scale-adjusted sample image, into the image model, wherein the object extraction component is used for extracting concerned objects in sample images at respective scales; inputs a feature of the scale-adjusted sample image into a local classifier in the image model respectively, performs category prediction with respect to feature points in the feature, so as to obtain a local prediction result, and updates the object extraction component based on the local prediction result; performs object level category prediction for the scale-adjusted sample image based on the feature and the updated object extraction component; and trains the image model based on a category prediction result of the scale-adjusted sample image.

Method and apparatus for training image model, and method and apparatus for category prediction

An image processing apparatus and an image processing method where the apparatus includes: a self-encoder configured to perform self-encoding on an input image to generate multiple feature maps; a parameter generator configured to generate multiple convolution kernels for a convolution neural network based on the multiple feature maps; and an outputter configured to generate, by using the convolution neural network, an output result of the input image based on the input image and the multiple convolution kernels. With the image processing apparatus and the image processing method according to the present disclosure, an accuracy of processing an image by using the CNN network can be improved.

Growing community of inventors

Beijing, China

Rujie Liu