Keypoint Separation Using Multi-Resolution Heat Maps and Vector Fields
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for object-specific keypoint separation face challenges in reducing computational load while maintaining accuracy, especially when handling high-resolution images.
Innovation Solution
The proposed solution involves an object-specific keypoint separation apparatus that uses a pre-trained model to output low-resolution heat maps and vector fields, which are then upscaled and corrected using vector density maps to maintain keypoint accuracy while reducing computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If heat maps and vector fields are generated with equal resolution to the input image, then keypoint estimation accuracy is improved, but computational load and memory requirements increase significantly
Solution Approach 1:
The patent divides the image processing into multiple resolution levels. Heat maps are generated at reduced resolution to decrease computational load, while vector fields are generated at full resolution to maintain keypoint estimation accuracy. This segmentation of processing tasks across different resolution levels resolves the contradiction between accuracy and computational complexity.
Solution Approach 2:
The patent applies different quality levels to different components: full-resolution vector fields are used locally at keypoint regions where high precision is critical, while heat maps use reduced resolution where lower precision is acceptable. This local quality differentiation maintains overall system accuracy while reducing global computational burden.
2Productivity
If the resolution of heat maps and vector fields is reduced, then computational load is decreased, but accuracy in estimation of keypoint coordinates deteriorates
Solution Approach 1:
The patent segments the resolution requirements for different data structures: heat maps use reduced resolution for faster processing, while vector fields maintain full resolution for accurate keypoint coordinate estimation. This segmentation allows the system to achieve both improved productivity and maintained precision.
Solution Approach 2:
The patent introduces an intermediary processing step where reduced-resolution heat maps are used for initial keypoint detection, and then full-resolution vector fields are applied to refine the keypoint coordinates. This intermediary approach allows efficient initial processing followed by precise refinement.
Data Source
AI summary
An object-specific keypoint separation apparatus includes: an inference execution unit configured to receive a captured image capturing an object as an input and use a pre-trained model that has been trained in order to output a plurality of first maps and a plurality of second maps generated from the input captured image to output the plurality of first maps and the plurality of second maps, the plurality of first maps storing a vector describing a connection relationship of a keypoint of the object only around the keypoint, and the plurality of second maps representing a heat map configured to have a peak at coordinates at which the keypoint of the object appears; a map correction unit configured to correct the plurality of second maps using the plurality of first maps and the plurality of second maps; an upsampling unit configured to upsample the plurality of first maps; and an object-specific keypoint separation unit configured to separate keypoints for each object based on the plurality of upsampled first maps and the plurality of corrected second maps.