Feature coding model, training method of visual relationship detection model and detection method

Abstract

The invention relates to the technical field of visual relation detection, in particular to a feature coding model, a training method of a visual relation detection model and a detection method. The training method of the feature coding model comprises the steps of obtaining an initial feature coding model; acquiring sample data; inputting each piece of sample data into the initial feature codingmodel; extracting a guide graph from the visual common sense data based on the category; and training the initial feature coding model according to the guide graph, and adjusting the conversion matrixto update the target feature of each target area to obtain the target feature code of each target area. On one hand, the defect of insufficient sample data is overcome by utilizing a guide graph related to the category in the visual common sense. Therefore, enough sample data support can be provided when the target features are coded again, on the other hand, it is guaranteed that relationship perception is introduced when the target features are coded, conditions are provided for subsequent visual relationship detection, and then the accuracy of visual relationship detection can be improved.

Description

technical field [0001] The invention relates to the technical field of visual relationship detection, in particular to a feature encoding model, a training method and a detection method for the visual relationship detection model. Background technique [0002] In recent years, deep learning has made breakthroughs in image recognition (such as image classification, object detection, object segmentation, etc.) tasks. Among them, to achieve computer understanding of the scene, an important part is visual relationship detection, that is, for an input picture, predict the position and category of the target object in the picture, as well as the category of the relationship between the target and the target. [0003] The commonly used method for visual relationship detection is to encode the target and relationship, and then predict the target category and relationship category through a classifier. These methods often use recurrent neural networks to gradually fuse regional feat...

AI Technical Summary

Problems solved by technology

[0005] In view of this, an embodiment of the present invention provides a feature encoding model, a training method and a de

Images