DPSSD: Dual-Path Single-Shot Detector.

Area of Science:

• Computer Vision
• Deep Learning
• Machine Learning

Background:

• Object detection is a critical computer vision task with applications in surveillance and autonomous driving.
• Existing methods face challenges in effectively extracting features for multi-scale object detection.

Purpose of the Study:

• To propose a novel dual-path multi-scale object detection paradigm.
• To optimize the multi-scale object detection problem by enhancing feature information extraction.

Main Methods:

• Developed a single-stage general object detection algorithm named Dual-Path Single-Shot Detector (DPSSD).
• Utilized a dual-path network (residual and concatenation paths) to leverage shallow features for improved accuracy.
• Integrated a feature fusion module to create a multi-scale feature learning paradigm called the Dual-Path Feature Pyramid.

Main Results:

• Trained and validated models on PASCAL VOC and COCO datasets with varying input resolutions (320 and 512 pixels).
• Demonstrated superior performance compared to anchor-based single-stage object detection algorithms.
• Achieved an advanced level of average accuracy in object detection tasks.

Conclusions:

• The proposed DPSSD algorithm effectively extracts abundant feature information for object detection.
• The dual-path network architecture enhances adaptability to multi-scale object detection challenges.
• The Dual-Path Feature Pyramid paradigm offers a robust approach for multi-scale feature learning, outperforming existing methods.