18results about How to "Reduce memory overhead" patented technology

The present disclosure provides a message processing method and device, equipment and storage medium, comprising: in response to receiving a to-be-processed message, extracting quintuple information from the to-be-processed message; matching the quintuple information with the index key of the session table to determine the type of the to-be-processed message, wherein the index key of the session table is constructed based on the quintuple information of the established session; in response to the type of the to-be-processed message being a first packet, matching the quintuple information with the index key of the association mapping table to determine whether the to-be-processed message is an associated session of the main session, wherein the index key of the association mapping table is constructed based on the IP address and port information obtained by parsing the application layer of the control channel message of the multi-channel protocol, and the main session is the session of the control channel; in response to the to-be-processed message being an associated session of the main session, performing forwarding processing on the to-be-processed message.

Embodiments of the present specification provide a method for question and answer reasoning of image-text content based on a target model, the target model comprising a plurality of decoding layers, and the image-text content comprising a plurality of images and a plurality of text segments; the method comprising: for a first image, performing first representation compression to obtain L visual compressed representations corresponding to the L decoding layers; for a first text segment, performing second representation compression to obtain L text compressed representations corresponding to the L decoding layers; replacing the images and the text segments in the image-text content with placeholders respectively, and inputting the plurality of decoding layers for decoding operation, comprising: in each decoding layer of the L decoding layers, superimposing the visual compressed representation and the text compressed representation corresponding to the decoding layer on the current hidden representation of the corresponding placeholder, and then performing fusion processing to obtain a decoding representation; and generating an answer text corresponding to the image-text content based on the decoding representation output by the last decoding layer.

The present disclosure relates to a layered parallel inference acceleration system, method and computer program product. The system comprises a text content analysis module for analyzing and processing user input text content to obtain a vector sequence; a candidate token sequence generation module for determining a corresponding set of to-be-predicted positions according to the current position of each of the plurality of vectors, and determining the feature information corresponding to each to-be-predicted position in the set of to-be-predicted positions; and predicting at least one candidate token according to the feature information, and combining each corresponding at least one candidate token in the set of to-be-predicted positions to generate a plurality of candidate token sequences; and a token sequence verification module for screening a candidate token sequence with the most number of candidate tokens that are continuously verified from the start position from the plurality of candidate token sequences as a target token sequence. In this way, the diversity and quality of the generated content can be ensured while reducing the model inference delay.

This invention discloses a message processing method and related apparatus, relating to the field of message processing. When the current message cache is not full, the main thread receives live messages and submits them to a HandlerThread for asynchronous parsing to obtain different message types. Corresponding caching strategies are then applied, retrieving image cache data from the ImageSpan cache and text cache data from the plain text cache, assembling them into a complete live message, and inserting it into a RecycleView queue for message aggregation caching. The complete live message is then inserted into a scrolling control in the RecycleView list for display. Based on a preset scrolling strategy, the complete live message is directly scrolled and displayed using a SpannableString cache. This invention utilizes HandlerThread for asynchronous message processing, a three-layer caching mechanism to optimize performance, and strategy parsing to decouple rendering logic, ensuring the main thread maintains a high frame rate.

This application discloses a method and apparatus for visualizing video instance segmentation masks. The method includes: receiving instance mask data from a server; parsing the compressed mask string character by character, recovering the corresponding run length count values ​​sequentially during the parsing process; writing the corresponding mask pixel value into a pixel buffer based on the currently recovered run length count value to obtain the corresponding pixel buffer data; generating the corresponding image data based on the pixel buffer data; and rendering the image data onto the browser canvas. This method, during the character-by-character parsing of the compressed mask string, directly writes the corresponding mask pixel value into the pixel buffer after recovering each run length count value, eliminating the need to store the complete run length count sequence during the decoding process. This reduces intermediate data memory usage, lowers memory overhead in the browser environment, and improves the smoothness of mask visualization rendering.

The invention belongs to the field of motors, and discloses a harmonic blocking type consequent-pole permanent magnet hub motor and a harmonic group differential vibration suppression method. The harmonic wave blocking type consequent pole permanent magnet hub motor provided by the invention adopts a harmonic wave trapping stator and a local magnetic short circuit magnetic barrier rotor structure under consequent poles to actively suppress the vibration source harmonic wave amplitude. The harmonic group differential vibration suppression method provided by the invention comprises the following steps: actively identifying a main vibration source under a current working condition, extracting a characteristic frequency of a superior harmonic group as a core optimization object, obtaining a Pareto leading edge and a solution set thereof under each working condition by adopting a multi-objective optimization method, taking an intersection of parameters of different working conditions, screening out a parameter set with optimal comprehensive performance, and carrying out differential vibration suppression on the Pareto leading edge and the solution set of the Pareto leading edge. The basic harmonic waves of the vibration source are classified into a monitoring layer, and monitoring temporary inactive compensation is carried out, so that differentiated accurate control is realized. According to the method, the limitation that a traditional method is fixed and single is overcome, and active suppression of the dynamic change main vibration source in the full working condition range of the consequent-pole permanent magnet hub motor is achieved.

The present application relates to a kind of data stream integrity detection method, device, equipment and medium, its method includes: receiving the source code to be detected input by user generates original assembly file with line number information;According to read instruction and write instruction in original assembly file, define sdDFIx instruction and ldDFIx instruction using the EUI attribute of preset data;Again, code insertion is carried out to original assembly file, and new assembly file is generated;Using original instruction set architecture, the new instruction set architecture of the new assembly file containing sdDFIx instruction and ldDFIx instruction after compiling is extended to execute;Using the new instruction set architecture after extension, the new assembly file after compiling and linking is executed, and the inspection result to the code to be checked is obtained;The purpose of the present application is to solve the defects that performance overhead is large in prior art, cannot more efficiently monitor program abnormal behavior, cannot effectively protect the problems of program runtime.

The invention discloses a picture generation method and device, a storage medium and equipment. The method comprises the steps of obtaining sparse voxel illumination data corresponding to a highest detail level model in a virtual scene; determining a cascade coordinate space corresponding to the virtual scene according to the spatial position and the observation depth of the virtual camera, and mapping the world coordinates of each detail hierarchy model into the cascade coordinate space; determining a boundary region of the adjacent detail hierarchy models in the cascade coordinate space; obtaining original voxel data of the current detail hierarchy model corresponding to the boundary region; the sparse voxel illumination data is loaded in the boundary region, and the sparse voxel illumination data and the original voxel data are mixed to generate a rendering picture of the virtual scene, so that smooth visual transition of the illumination effect between adjacent detail hierarchy models is realized, and the memory consumption is effectively reduced.

The application discloses a multi-modal three-dimensional target detection method based on structural feature and semantic feature fusion, and belongs to the technical field of target detection.The application solves the problems of low precision, poor real-time performance and insufficient robustness of the existing method.The application constructs an explicit matching fusion mechanism of low-layer structure guided initialization, high-layer semantic optimization and time sequence expansion, projects 3D boundary boxes of a laser radar and a camera to multi-view image planes, calculates geometric similarity and category consistency constraints of 2D projection regions, and constructs an explicit matching graph in combination with low-layer structure information and high-layer semantic features.The application guides weighted aggregation of cross-modal target features with high confidence through a sparse matching graph, aligns historical frame targets to a current frame coordinate system to construct a space-time matching graph, aggregates historical information through expansion to a time sequence dimension, and realizes efficient and accurate three-dimensional target detection of the laser radar and the visual multi-view camera.The application method can be applied to multi-modal three-dimensional target detection.

The application provides a log collection method, system, device and medium based on an MQTT protocol, relates to the technical field of data collection, and comprises the following steps: acquiring a plurality of original log data; storing each original log data to a ring buffer area according to a non-blocking coverage strategy, wherein the ring buffer area comprises a preset number of continuous storage blocks; reading a plurality of target log data from the ring buffer area; performing hierarchical filtering on each target log data according to a preset filtering rule to obtain a plurality of filtered log data; performing binary encoding compression on each filtered log data to obtain a plurality of compressed log data; storing each compressed log data in a preset memory pool based on a dynamic memory allocation strategy; and transmitting each compressed log data in the preset memory pool to a remote server based on a lightweight processed MQTT protocol. The application reduces the memory overhead required for log collection while improving the log transmission efficiency and transmission stability.

The invention discloses a feature reconstruction method and device based on Token scores, and belongs to the technical field of computer science and artificial intelligence. The method comprises the following steps: performing word segmentation processing on a binary file to obtain a plurality of Tokens; obtaining a text feature and a visual feature of each Token; the text feature is obtained by mapping the Token into a unique ID and then carrying out vector conversion; the visual features are extracted based on a coded image converted from a binary file; performing importance scoring on the corresponding Token by using the text feature and the visual feature of each Token to obtain a score of each Token; calculating attention weights for the K Tokens with the highest scores, and performing feature weighted fusion on the K Tokens with the highest scores according to the attention weights to obtain attention features; performing feature reconstruction on the attention features and the pooling features to perform task processing by using the reconstructed features; the pooling feature is obtained by performing pooling fusion on the features of the remaining Token. According to the invention, the accuracy of the detection result can be ensured while the calculation complexity is reduced.

The invention relates to the technical field of data analysis, can be applied to business system platforms of financial science and technology, medical health and the like, and discloses a prototype memory enhanced video key frame extraction method, device, equipment and medium, the method comprises the following steps: carrying out time sequence pulse convolution on each video frame in a time sequence pulse sequence converted from original video data; performing correlation matching on the pulse feature sequence and differentiable prototype features in a preset prototype memory bank to obtain a related memory feature sequence, and performing dynamic fusion on the pulse feature sequence and the related memory feature sequence to obtain memory enhanced pulse features; and performing key analysis on the video frames with the memory enhanced pulse features to obtain key frame discrimination scores, and screening out video key frames. According to the method, overfitting is effectively suppressed in a few-sample scene, video dynamic time sequence information is accurately captured, and the efficiency and accuracy of key frame recognition are improved while the calculation complexity and memory overhead are reduced.

The application provides a user environment scene recognition method, system and electronic equipment, comprising: sequentially completing multi-modal original data acquisition and alignment, noise reduction and normalization, multi-scale feature extraction, time sequence structure decision and dynamic network structure adjustment, through constructing time sequence consistency gating mechanism and dynamic structure interpolation control signal, realizing adaptive smooth adjustment of channel activation number and attention head number of a lightweight mixed backbone network, and then driving a learnable soft mask to sparsely schedule backbone network level calculation on demand, finally completing efficient and accurate environment scene recognition, the application effectively improves the calculation efficiency and inference accuracy of the scene recognition model under the condition that the terminal device resources are limited, and has good structure continuity and hardware adaptability.

The invention relates to the technical field of image processing, in particular to a feedforward type large-scale scene reconstruction method based on three-dimensional Gaussian splashing. The method comprises the following steps: acquiring a multi-view image set and constructing an initial camera token carrying view angle identification information; carrying out feature interaction between the intra-frame and the cross-view through an alternating attention geometry Transform, and outputting an updated camera token and an updated image feature token; the pose, the depth map and the depth confidence map of each frame are predicted through decoding branches; the pixels are projected back to a three-dimensional space by using the prediction parameters, and Gaussian attributes and confidence coefficients are regressed by combining image features; and finally, realizing redundancy elimination of Gaussian primitives through voxelization aggregation, and outputting a target view. According to the method, high-efficiency and high-precision three-dimensional reconstruction of a large scene can be realized without the prior of internal and external parameters of a camera and a complex sparse reconstruction process, the calculation overhead is remarkably reduced, and the problems of geometric drift and video memory overflow which are easy to occur in the large scene in a traditional method are solved.

This application discloses a memory management method and electronic device, relating to the field of big data technology. It includes generating a first interface for managing available user-mode memory, enabling user-mode sharing of multi-level page tables; and using a second interface to pre-register data transmission regions for multiple access devices and generate a first access key. When an access device is invoked to perform data transmission / reception, the first interface accesses the multi-level page table, and the virtual address carried in the data transmission / reception request is converted to a physical address based on the multi-level page table. The physical address is then transmitted to the access device, allowing the access device to perform data transmission / reception operations on the memory region corresponding to the physical address based on its matching first access key and the physical address. Relying on a unified multi-level page table shared by multiple access devices effectively reduces the memory overhead of traditional mapping tables. The pre-registration via the second interface effectively solves the problem of high latency in registering memory regions during memory transmission.

The invention relates to the technical field of distributed cache and memory computing, and discloses an SLRU page replacement algorithm optimization method based on Apache Ignite, and the method comprises the steps: constructing an improved SLRU data structure comprising a trial segment and a protection segment, and a page heat table based on an array index; the page access request is responded to update the popularity, access judgment based on a popularity threshold value is executed, and the page popularity is promoted to a protection section only when the page popularity of a trial section meets an access condition; monitoring the page view and the time interval, triggering dynamic heat attenuation, calculating an attenuation coefficient according to the hit rate of the protection section, and updating the total heat; monitoring the hit rate of the protection section, calculating the capacity expansion and contraction proportion, and dynamically adjusting the capacity proportion of the trial section and the protection section. Through the independent popularity model, the strict admission mechanism and the self-adaptive adjustment strategy, the cache pollution is effectively reduced, the problem of historical popularity lag is solved, and the cache hit rate and the resource utilization rate in a high-concurrency environment are improved.

This invention belongs to the field of caching and discloses a key-value cache compression method and apparatus for anomaly detection in power distribution network equipment. The method includes: extracting patch features from images of power distribution network equipment using a multimodal large-scale visual encoder, and generating an image anomaly heatmap reflecting defect distribution by comparing features with those of normal samples of the same category; constructing an anomaly importance field with semantic boundary isolation based on the structural semantic boundaries and distribution information of anomaly regions in the images of power distribution network equipment; dividing the image patch network into multiple non-overlapping feature blocks, adaptively allocating the total number of key-value cache tokens for each image of power distribution network equipment, and differentially allocating token retention quotas according to the importance weights of each feature block; filtering and retaining visual tokens according to importance priority, and simplifying text tokens by combining anchor point forced retention with general content downsampling; and fusing the filtered visual and text tokens to generate a compressed key-value cache.

The invention provides a binary code instrumentation method, system and equipment, a storage medium and a program product. The binary code instrumentation method can be applied to a combined processing device. The combined processing device comprises a computing device, an interface device, a processing device and a storage device. The computing device is configured to execute a dynamic binary instrumentation task, is mainly implemented as a single-core intelligent processor or a multi-core intelligent processor, is used for executing deep learning or machine learning computation, and can interact with the processing device through the interface device to jointly complete the dynamic binary instrumentation task. The interface device is used for transmitting data and control instructions between the computing device and the processing device. As a general processing device, the processing device performs basic control including but not limited to data handling, starting and / or stopping of the computing device, and the like. The storage device is used for storing data to be processed. According to the scheme, the flexibility of dynamic binary instrumentation is improved, and the application range is expanded.