Keypoint Positioning via Attention Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current deep learning methods for keypoint positioning in images often fail to accurately locate keypoints due to shielding effects from location, viewing angle, or other factors, resulting in low positioning accuracy.
Innovation Solution
An image data processing method that involves obtaining a source image with a target object, extracting a local feature sequence, performing location encoding to obtain location encoding information, combining the local feature sequence with this information to form an object description feature, and using an attention output feature to represent the information transfer relationship between global features, ultimately determining the keypoint location information based on the object encoding feature.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If deep learning model extracts local feature for keypoint positioning, then positioning process can be performed, but positioning accuracy deteriorates due to shielded parts not being described in local feature
Solution Approach 1:
The patent segments the feature extraction process into local feature extraction (using deep learning models) and global feature consideration (using attention mechanisms). The local feature sequence captures detailed information from visible parts, while the attention output feature captures relationships between global features. This segmentation allows the system to handle shielded parts by considering global context rather than relying solely on local visible features.
Solution Approach 2:
The patent introduces an intermediary mechanism - the attention encoding component - that processes the local feature sequence and generates an attention output feature representing information transfer relationships between global features. This intermediary bridges the gap between local extracted features and global object understanding, enabling the system to infer characteristics of shielded parts from visible parts through attention-based relationships.
2Productivity
If only local feature is used for keypoint positioning, then processing speed is fast, but positioning accuracy is low due to lack of global context
Solution Approach 1:
The patent merges two processing paths: the local feature extraction path (fast deep learning inference) and the global feature relationship path (attention mechanism). The local feature sequence is processed through location encoding to get location encoding information, which is then combined with the original local features. The attention output feature representing global relationships is also integrated. This merging allows the system to maintain processing speed while improving accuracy through comprehensive feature utilization.
Solution Approach 2:
The patent adds another dimension to the feature representation by introducing the attention output feature that captures relationships between global features. Instead of only using local features in the original dimension, the system now operates in an enhanced feature space that includes both local features and global relationship information. This dimensional expansion enables the system to leverage global context without significantly increasing computational cost.
Data Source
AI summary
Techniques for image data processing and image detected are described herein. Techniques may include obtaining a source image that includes a target object, and obtaining a local feature sequence of the target object from the source image; performing location encoding processing on the local feature sequence to obtain location encoding information of the local feature sequence, and combining the local feature sequence and the location encoding information into an object description feature associated with the target object; obtaining an attention output feature of the object description feature; the attention output feature being configured for representing an information transfer relationship between global features of the target object; and determining an object encoding feature of the source image according to the object description feature and the attention output feature; and determining keypoint location information of the target object in the source image based on the object encoding feature.