Anchor-Free Object Detection via Keypoint Heat Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods, such as R-CNN, Fast R-CNN, and Faster R-CNN, generate excessive bounding boxes for each object, leading to increased computational requirements and slower detection performance.
Innovation Solution
A computing device employing an anchor-free method generates region information for objects in an image by using a key point heat map to determine key points of objects, then generates non-rotating and rotating bounding boxes based on this information, reducing the need for excessive bounding box generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If anchor-based techniques generate multiple bounding boxes for each object, then object detection coverage is improved, but computational complexity increases and detection speed decreases
Solution Approach 1:
The patent extracts the essential function of anchor-based detection by removing the anchor generation step entirely. Instead of generating multiple candidate bounding boxes around anchor points, the model directly predicts bounding box parameters from keypoint detections, eliminating redundant computation while preserving detection coverage through learned keypoint-to-box mappings.
Solution Approach 2:
The patent changes the parameter representation from anchor-based candidate boxes to keypoint-based direct predictions. By representing objects through keypoint coordinates and directly predicting bounding box parameters from these keypoints, the system transforms the detection paradigm to reduce computational parameters while maintaining detection accuracy.
2Measurement precision
If anchor-based techniques generate multiple bounding boxes for each object, then detection accuracy is maintained, but the amount of calculation increases
Solution Approach 1:
The patent extracts and removes the computationally intensive anchor generation and selection process, keeping only the essential function of predicting accurate bounding boxes. By directly regressing bounding box parameters from keypoint features, the system achieves comparable detection accuracy with significantly reduced computational power requirements.
Solution Approach 2:
The patent substitutes the mechanical anchor-based selection process with a learned direct prediction mechanism. Instead of generating multiple candidate boxes and selecting the best match through comparison, the system uses neural network regression to directly predict optimal bounding box parameters from keypoint representations, replacing iterative mechanical selection with efficient direct computation.
Data Source
AI summary
Disclosed is a computing device for generating region information of at least one object included in an image, the computing device including: a memory including computer-executable components; and a processor for executing the computer-executable components stored in the memory, in which the computer-executable components include: a key point heat map generation component for generating, for the image, a key point heat map including key point information of the at least one object; a non-rotating bounding box generation component for generating a non-rotating bounding box based on the generated key point heat map; a rotating bounding box generation component for generating a rotating bounding box by using the non-rotating bounding box; and a final bounding box generation component for representing a region occupied by the at least one object in the image by using at least one of the non-rotating bounding box or the rotating bounding box.