Related Experiment Video
Updated: Oct 13, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
682
SEFPN: Scale-Equalizing Feature Pyramid Network for Object Detection
Zhiqiang Zhang1,2, Xin Qiu1, Yongzhou Li1
1Institute of Microelectronics, Chinese Academy of Sciences, No. 3 Beitucheng West Road, Chaoyang District, Beijing 100029, China.
Sensors (Basel, Switzerland)
|November 13, 2021
Summary
Feature Pyramid Network (FPN) defects are addressed by the Scale-Equalizing Feature Pyramid Network (SEFPN). SEFPN balances feature representation across levels, improving object detection performance in one-stage and two-stage networks, especially for large objects.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Feature Pyramid Network (FPN) is a common component in object detection, but suffers from information loss and semantic gaps between feature levels.
- Existing methods like correlation convolution only partially address imbalances between adjacent layers.
Purpose of the Study:
- To propose a novel network structure, Scale-Equalizing Feature Pyramid Network (SEFPN), to overcome the limitations of FPN.
- To enhance semantic representation balance across all feature levels in object detection.
Main Methods:
- Introduced SEFPN, a network structure that iteratively fuses features from each level to generate multi-scale features.
- Integrated SEFPN as a standalone module into existing object detection frameworks.
Main Results:
- SEFPN integration improved one-stage detectors by approximately 1 AP (Average Precision).
- SEFPN enhanced Faster R-CNN performance, particularly for large object detection (APL increased by ~2 AP).
Conclusions:
- SEFPN effectively balances semantic representation across feature levels, mitigating FPN's inherent defects.
- The proposed SEFPN offers a simple yet effective method to boost object detection accuracy, especially for large objects.
