一种基于frustum deformable attention的目标检测方法
By introducing Frustum Deformable Attention into the Deformable Attention model, the sampling strategy of the feature layer is optimized, solving the problems of computational waste and GPU memory waste, and achieving more efficient object detection and more accurate detection results.
CN117392426BActive Publication Date: 2026-07-17BRETON TECHNOLOGY CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BRETON TECHNOLOGY CO LTD
- Filing Date
- 2023-08-28
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing Deformable Attention models suffer from computational and memory waste in object detection, especially due to indiscriminate sampling across different feature layers, leading to computational chaos and inefficiency.
Method used
The Frustum Deformable Attention method is adopted, which reduces invalid computation and improves sampling efficiency by setting different sampling points and attention weights on different feature layers.
Benefits of technology
It improves the efficiency and accuracy of object detection, reduces computational and spatial complexity, and generates multi-scale feature maps with more complete semantics, thereby enhancing detection performance.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN117392426B_ABST
Abstract
本发明的一种基于frustum deformable attention的目标检测方法，属于深度学习目标检测技术领域。所述方法采用步骤如下：S100、输入数据；S200、特征提取；S300、模型处理；S400、输出目标检测预测值。本发明的模型基于frustum deformable attention进行模型训练后，其结构输出的四层多尺度特征图，在依然包含自身层的含义和表示功能的同时，底层纹理形成了更加贴近描绘目标的纹理层，高层抽象特征结合了局部纹理和整体结构从而形成了更具完整语义的表示信息。多尺度特征依图在保持层次分明的同时，在自身包含的信息维度内形成了对画面目标更有力的捕捉。
Need to check novelty before this filing date? Find Prior Art