100 results about "Feature sharing" patented technology

The invention discloses a method for 3D reconstruction and texture generation from single-view face based on multi-task learning, belonging to the computer vision field. The method comprises the following steps of: selecting a special viewpoint for rendering a three-dimensional human face model; generating depth map and texture map as truth value data from special viewpoint; designing an integrated learning and coding network based on feature sharing of depth information and texture information; designing the branch decoding network to recover the depth map from the shared features, and recovering the depth map, designing the mutual information with shared features as latent variables to maximize the generation of antagonistic network, and restoring the texture unwrapping map, adjusting the proportion of each task loss function and training the model; interpolating the depth map and using texture map to reconstruct the 3D face mesh model with texture details. The invention utilizes multi-task learning to carry out single-view face three-dimensional reconstruction, texture generation and style migration, and has the advantages of high speed, low cost and the like.

The invention discloses a bill image text detection and recognition method based on deep learning, and the method comprises the steps of carrying out the feature extraction on a bill image through employing a convolutional neural network, and generating a first feature map; and then using the text detection network to perform the multi-task prediction on the first feature map through classification and regression operation to obtain a detection textbox; and on this basis, using the text recognition network to map the detection text box to an area corresponding to the first feature map and perform pooling operation; generating a second feature map with the fixed height and width changing in proportion, converting the second feature map into a feature sequence, coding the context informationof the feature sequence by adopting a recurrent neural network, and then decoding by adopting a group of recurrent neural networks with attention mechanisms to obtain a recognition result of the textarea. According to the present invention, the detection task and the identification task are integrated into a unified network framework, the convolutional layer feature sharing is achieved, the end-to-end joint training can be completed, and the overall identification performance of the model is improved.

A method of aggregating events in a PCIe (Peripheral Component Interconnect Express) multifunction device minimizes reported error messages, where several functions share a common PCIe interface logic. A predetermined number of function entities with logical gates, connected in daisy chain configuration, process incoming information, and a decision is made whether each function entity will generate a blocking control or a pass-through control. The error messages are aggregated across the function entities in a single clock cycle with the help of an error controller. The functions can be from IEEE 1394 interface, graphics display controller, sound card, PCIe switch, or PCIe to PCI bridge connection. Each function preferably has a different configuration and security level setting for error reporting and messaging. There may be a plurality of parallel daisy chains, and the PCIe device may include three layers namely, a physical layer, data link layer and transaction protocol layer (for error logging, reporting).

According to one aspect of the present invention, there is provided a mobile telephone operable to use a feature of a second mobile telephone comprising a share module for obtaining details of one or more features on the second mobile telephone which are available for use by the mobile telephone a user interface for selecting an available feature; and a communication module for communicating data with the second mobile telephone to enable the selected feature to be used by the mobile telephone.

Aggregated service and feature sharing is provided in a home network, in which users can freely enjoy not only the shared content, but also the unique features provided by each device anywhere in the network. The service sharing feature allows one device to control and use shared hardware or software functionalities provided by other devices in the network.

The invention provides a cloud terminal platform integrating application systems and an integrating method, and discloses an information system integrating platform. The information system integrating platform comprises a server side and a client side, wherein the server side comprises a data module, an application module, and an interface unit communicating data with the client side; the data module comprises a data center unit and a data service unit; the application module comprises an application core integrating unit, an application interface integrating unit, a management core integrating unit and a unified identity certification unit; the client side comprises an application integrating unit, an interface integrating unit and an application service unit. The information system integrating platform disclosed by the invention effectively integrates all business application systems, and realizes not only integrated display of a B/S framework system and a C/S framework system, but also functions such as information sharing, function sharing and interface sharing, and inter-system team working.

One or more virtual functions are exposed via a shared communication interface. Memory across said virtual functions is shared to provide a fixed number of I/O buffers shared across said virtual functions. For each of said one or more virtual functions, storing a corresponding map table configured to store a mapping data that maps a logical block address of the virtual function to a corresponding allocated one of said fixed number of I/O buffers based at least in part on a current state of a state machine.

A system includes a virtualized I/O device coupled to one or more processing units. The virtualized I/O device includes programmed I/O (PIO) configuration registers corresponding to hardware resources, and a storage for storing a resource table that includes a plurality of entries. Each entry corresponds to a respective hardware resource. A system processor may allocate the hardware resources to functions that may include physical and virtual functions, and may program each entry of the resource discovery table for each function with an encoded value that indicates whether a requested hardware resource has been allocated to a requesting process, and whether the requested hardware resource is shared with another function. Processing units may execute a device driver instance associated with a given process to discover allocated resources by requesting access to the resource discovery table. The virtualized I/O device protects the resources by checking access requests against the resource discovery table.

The invention relates to the technical field of computers, in particular to a guarantee method and a guarantee system for improving usability of virtual machine application. The system realizes communications by means of an event communication mechanism for performing communications, and mainly comprises a shared data base, a communication module, a service perceiving module, a service strategy module, a service execution module and a virtue machine management module. The service perceiving module is used for perceiving application state; the service strategy utilizes the state perceived by the service perceiving module for selecting high availability strategy; the service execution module transfers in turn related functions of the virtue machine management module in corresponding orders according to the high availability strategy; and the shared data base provides information required in operation for the modules. The guarantee method and the guarantee system for improving usability of virtual machine application have the advantages of being capable of providing sustained security guarantee for application business (operational procedure installed to operate in a virtue machine) of the virtue machine, and guaranteeing stable and reliable operation of application services.

The invention relates to an image analysis method and device. The method is implemented through an analysis model. The model comprises a feature sharing module, a semantic segmentation module and an edge detection module. The method comprises the steps that feature extraction processing is carried out on an to-be-analyzed image through the feature sharing module, and shared features are acquired,wherein the shared features comprise feature information of a plurality of network depths; semantic segmentation processing and edge detection processing are carried out on the shared features throughthe semantic segmentation module and the edge detection module respectively, and a preliminary semantic segmentation result and a preliminary edge detection result of the to-be-analyzed image are obtained. According to the method and device, the shared features of the to-be-analyzed image can be extracted through the analysis model, and semantic segmentation processing and edge detection processing are carried out on the shared features to obtain the preliminary semantic segmentation result and the preliminary edge detection result, so that the consistency between a semantic segmentation result and an edge detection result is improved.

The invention relates to a filtering and shunting device and a method supporting a multi-service function. The method comprises the following steps of: matching input original data packets with regular policy library subunits established by different users one by one to form match results; processing the match results to form result labels, and establishing a mapping table among the result labels and output ports; and adding the result labels to the heads of the original data packets, obtaining the output ports according to the result labels and the mapping table, and sending the original data packets to users from the output ports. The shunting and filtering functions of a plurality of services share the processing resources in the same input unit and device, a plurality of sets of filtering and shunting policies are supported in a set of filtering and shunting devices, are completely independent, and do not interfered one another, and the requirement of the multi-service function is effectively solved.

The invention provides a convolutional neural network-based target tracking method and apparatus. The method comprises the steps of constructing a convolutional neural network, wherein the convolutional neural network comprises a plurality of feature sharing layers and a specific classification layer; performing offline training: for different video sequences with target labels, performing training by adopting different specific classification layers and the feature sharing layers, and updating network parameters, until the network is converged or a preset iterative frequency is reached, wherein the network parameters include parameters of the specific classification layers and parameters of the feature sharing layers; and updating the convolutional neural network, obtaining an optimal region from a video frame by using the convolutional neural network, and setting the optimal region as a tracking region. According to the method and the apparatus, the deficiency of artificial features is overcome, the defect of a general convolutional neural network in a tracking problem is improved, and the robustness of a target tracking algorithm is improved.

The invention provides a method and a system for combined optical performance monitoring, rate and modulation pattern recognition, which can identify optical signal rate and modulation format while realize optical performance monitoring. The method comprises the following steps of: acquiring two-dimensional scatter maps of different rates and different modulation formats under different light performance damages, and dividing the acquired two-dimensional scatter maps into a training set and a test set; A multi-task convolutional neural network model is constructed, in which the first four layers are convolutional pooling layers, the fifth layer is feature sharing layer, and the sixth layer has three output ports, the first output port is optical performance damage monitoring output, the second output port is modulation format class output, and the third output port is rate class output. The training set is used as the input of the multi-task convolutional neural network model, and thecorresponding light performance damage value, rate category and modulation format category are used as the label value training model. The present invention is suitable for optical performance monitoring and rate/modulation format recognition operations.

The invention provides a fog computing intrusion detection feature sharing system based on a block chain technology, which can ensure the security of content of an intrusion detection feature libraryand is beneficial to improving the intrusion detection accuracy. The system comprises an identity authentication module used for carrying out identity authentication on a node to be added into an alliance chain, and if the authentication is passed, allowing the node to be added into the alliance chain; a feature sharing module used for broadcasting any node A added into the alliance chain to a network for sharing if the node A shares the updated intrusion detection feature library; and a feature verification module used for carrying out correctness verification on the intrusion detection feature library broadcasted in the network by other nodes in the network, and if the verification is passed, storing the intrusion detection feature library in a block chain system in an uplink manner. Theinvention relates to the technical field of information security.

The invention provides a planar map symbol model based on a path. The model comprises two parts of a planar map symbol data model and planar map symbol rendering. The invention also provides a planar map symbol sharing method based on the model. The method comprises a data sharing method and a function sharing method and comprises planar map symbol assimilation, planar map symbol XML exchange and planar symbol monitoring and replacement. The invention separates planar map symbol data and planar map symbol rendering and can assimilate a planar symbol manufactured by the current CAD, GIS and professional map software, support the design of a complex planar symbol and a cartoon symbol and texture filling and expand the planar symbol based on the path and the rendering thereof to a corresponding GIS software platform, thereby realizing the planar symbol sharing among the GIS, the CAD and the professional map software and providing a path for a map symbol library with multiple purposes.

The invention relates to internet vehicular equipment, in particular to intelligent integrated internet vehicular equipment which integrates multimedia playing, driving data recording, navigation and instant sharing of information, and a working module set of the intelligent integrated internet vehicular equipment. The intelligent integrated internet vehicular equipment mainly comprises a shell, a display screen, a control device and a power source, wherein the shell is of a two-piece combined shell structure, wherein a touch display screen is arranged in the front half shell in a nested manner; the control device mainly comprises a circuit control key, a working module group and a microprocessor; in the meantime, the control device is electrically connected with a camera device and a storage device, wherein the storage device is used for storing image information acquired by the camera device in real time. The intelligent integrated internet vehicular equipment integrates a multimedia player, an information sharing terminal, an automobile data recorder and a navigator, realizes the purposes of safety recording, real-time navigation, multimedia transmission in a mobile terminal and the like in a driving process and can share the expressed functions to other vehicular equipment, so that automobile owners can enjoy freer and rapider entertainment and information sharing, and map navigation sharing.

The invention provides an image instance segmentation method and device. The method comprises the steps of obtaining feature maps of multiple levels of an image; fusing at least two levels of featuremaps in the plurality of levels of feature maps to obtain a detection shared feature map; inputting the detection sharing feature map into a target detection sub-network to obtain a target area, the target being an object in the image; fusing the detection sharing feature map and the feature map of at least one level in the feature maps of the plurality of levels to obtain a segmentation sharing feature map; and segmenting the image in the target area according to the segmentation sharing feature map to obtain an instance segmentation result. According to the embodiment of the invention, feature sharing during target detection and target segmentation is improved, so that information propagation between the target detection task and the segmentation task is more sufficient, and the speed and precision of instance segmentation are improved.

The invention relates to a comprehensive energy load prediction method based on a multi-task learning strategy and deep learning. The method comprises the following steps: S1, forming a historical cooling, heating and power load feature library and an influence factor feature library; S2, constructing an MMoE multi-task learning model; S3, constructing three LSTM neural network models, training the three LSTM neural network models by utilizing feature sharing and extraction results output by a plurality of expert sub-networks in the MMoE model in the step S2 and cold, hot and electric load tags, and finally obtaining an MMoE-LSTM model; and S4, obtaining a cooling, heating and electric load prediction result. According to the method, interference of weak correlation information on the subtasks can be avoided, and a better prediction effect is achieved.

The invention relates to a content feature query method and device, computer equipment and a storage medium. The method comprises the steps of acquiring a content feature query request for target contents; determining a recommendation system logic layer initiating the content feature query request; obtaining a content feature category corresponding to the recommendation system logic layer; determining a feature sharing memory used for local persistent storage and corresponding to the content feature category; and from the determined feature sharing memory, querying content features of the target contents. According to the scheme, the query efficiency of the content features is improved.

The invention relates to a computer vision technology, in particular to an object attribute prediction method, device and equipment, and an object attribute prediction model construction method, device and equipment, and aims to solve the problems of relatively low prediction accuracy and relatively high calculation resource consumption in the prior art. All attributes are put together for featureextraction by using a shared feature extraction network; after the feature extraction network, according to the division standards of the ordered attributes and the disordered attributes, different types of attributes are placed in attribute prediction networks of a classification structure and a regression structure for processing, and different loss functions are adopted by the attribute prediction networks of different structures. According to the scheme provided by the invention, the correlation between the attributes is fully utilized, efficient multi-attribute feature sharing is realized, more effective features are mined out, and the heterogeneity between the attributes is fully utilized to realize attribute prediction for separately managing the ordered attributes and the disordered attributes, so that the prediction accuracy can be improved, and the computing resources are saved.

The invention relates to a lightweight image super-resolution reconstruction method based on a progressive distillation network, and relates to the technical field of image super-resolution reconstruction, and the method comprises the steps: carrying out the improvement of an existing super-resolution convolutional neural network model, including the steps: according to the receptive field progressive principle of convolution kernels with different expansion rates, utilizing two different progressive distillation connection combinations to replace original feature distillation connection, and adopting an asymmetric expansion convolution residual block, so the network can fully extract edge and texture feature information of an image under the condition of extremely few parameters; utilizing a channel shuffling structure to improve hierarchical features of distillation network connection, and further improving feature sharing among channels, so the accuracy of image super-resolution reconstruction is improved; further adopting a multi-scale space attention mechanism module, so the weight of the fusion features can be recalibrated in a self-adaptive mode; wherein the post-up-sampling reconstruction part is an up-sampling-based three-dimensional pixel attention mechanism method, and the efficiency of super-resolution image reconstruction is further improved.