Video Fine Granularity Recognition via Feature Graph Attention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video fine granularity recognition methods struggle to accurately distinguish between subtle differences in targets, resulting in lower recognition accuracy due to the inability to effectively extract and process detailed features from video frames.
Innovation Solution
A method involving sampling processing to obtain multiple frames from video, extracting feature graphs using deep neural networks, generating attention-weighted intermediate summary features, and performing redundant texture restraint to enhance feature extraction and classification, thereby improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If search-based video fine granularity recognition is used to extract fundamental attributes, then the recognition process is simple, but the accuracy of recognition result is low due to inability to extract subtle differences
Solution Approach 1:
The video is divided into multiple frames through sampling processing, and each frame is processed independently to extract feature graphs. This segmentation allows the system to capture subtle differences across multiple temporal instances, improving recognition accuracy while managing complexity through parallel processing of individual frames
Solution Approach 2:
The patent transitions from extracting simple fundamental attributes to extracting feature graphs that include subtle differences. By adding the dimension of detailed feature representation (from basic attributes to comprehensive feature graphs including color, shape, texture, and subtle differences), the system achieves higher recognition accuracy
2Measurement precision
If multiple frames are processed to extract detailed features, then recognition accuracy improves, but processing time increases
Solution Approach 1:
A deep neural network model is pre-trained beforehand to enable efficient feature graph extraction during actual video recognition. This preliminary training allows the system to quickly process multiple frames and extract detailed features without excessive processing time during inference
Solution Approach 2:
The patent combines feature graphs from multiple frames through summary feature determination, merging information across temporal instances. This consolidation allows the system to leverage detailed features from multiple frames while producing a single integrated recognition result, balancing accuracy improvement with processing efficiency
Data Source
AI summary
The present disclosure provides a method and apparatus for recognizing video fine granularity, a computer device and a storage medium, wherein the method comprises: performing sampling processing for video to be recognized to obtain n frames of images, n being a positive integer larger than one; respectively obtaining a feature graph of each frame of image, and determining a summary feature according to respective feature graphs; determining a fine granularity recognition result of a target in the video according to the summary feature. The solution of the present disclosure may be applied to enhance the accuracy of recognition result.