Dual-scale parallel training method for target detection model
By using a dual-scale parallel training method, high- and low-resolution images are generated in parallel to input the model for loss optimization. This solves the problems of high computational overhead and low efficiency caused by scale changes in existing technologies, and achieves efficient multi-scale target detection and cross-scale generalization capabilities.
CN121600359APending Publication Date: 2026-03-03GUILIN UNIV OF AEROSPACE TECH
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202511831028.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-06
- Publication Date
- 2026-03-03
Smart Images
Figure CN121600359A_ABST
Abstract
The invention discloses a dual-scale parallel training method for a target detection model, and belongs to the technical field of computer vision and deep learning. In order to solve the problems in the prior art that a model reasoning structure is difficult to change and reasoning time consumption is not increased while the model scale generalization ability is difficult to improve, cross-scale feature learning is realized by parallelly inputting a second resolution version of a homologous image in a training stage and fusing the loss of the second resolution version. According to the method, the multi-scale detection precision and the cross-scale generalization ability of the target detection model are remarkably improved.
Need to check novelty before this filing date? Find Prior Art