FGAA-FPN: Foreground-Guided Angle-Aware Feature Pyramid Network for Oriented Object Detection

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

FGAA-FPN, a Foreground-Guided Angle-Aware Feature Pyramid Network, is proposed for oriented object detection in high-resolution remote sensing imagery. This novel framework addresses challenges like cluttered backgrounds, scale variation, and large orientation changes by integrating explicit foreground modeling and geometric orientation priors into multi-scale feature fusion. FGAA-FPN employs a hierarchical design: a Foreground-Guided Feature Modulation (FGFM) module enhances object regions and suppresses background interference in low-level features (P3-P5), while an Angle-Aware Multi-Head Attention (AAMHA) module encodes relative orientation relationships for global interactions among high-level semantic features (P5-P7). Experiments on DOTA v1.0 and DOTA v1.5 datasets, using a ResNet-50 backbone and trained for 12 epochs on an NVIDIA RTX 5090 GPU with 1024x1024 images, demonstrate state-of-the-art results, achieving 75.5% and 68.3% mAP, respectively. The design also shows plug-and-play applicability across various detectors.

Key takeaway

For Machine Learning Engineers developing oriented object detection systems for remote sensing, consider adopting FGAA-FPN's hierarchical approach. You should integrate foreground-guided modulation at lower feature pyramid levels to suppress background noise. Simultaneously, apply angle-aware attention at higher levels to preserve directional structures. This strategy significantly boosts mAP on DOTA v1.0 and v1.5, offering a plug-and-play enhancement for existing two-stage detectors. Be aware that FGAA-FPN introduces additional computational cost.

Key insights

Hierarchical foreground-guided and angle-aware feature modulation improves oriented object detection in complex remote sensing imagery.

Principles

• Differentiate feature processing across pyramid levels.
• Explicitly model foreground saliency for low-level features.
• Encode relative orientation for high-level semantic features.

Method

FGAA-FPN integrates Foreground-Guided Feature Modulation (FGFM) at lower pyramid levels (P3-P5) and Angle-Aware Multi-Head Attention (AAMHA) at higher levels (P5-P7) into a two-stage oriented detector, using weakly supervised foreground loss.

In practice

• Apply FGFM to P3-P5 for high-resolution features.
• Deploy AAMHA on P5-P7 for cost-effective orientation modeling.
• Use both orientation and mask biases in AAMHA for robustness.

Topics

• Oriented Object Detection
• Feature Pyramid Networks
• Remote Sensing Imagery
• Multi-scale Feature Fusion
• Attention Mechanisms
• Foreground-Guided Modulation
• Angle-Aware Attention

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• FPN Paper Walkthrough: Leveraging the Internal Pyramid — FPN enriches multi-scale feature maps with both high spatial and semantic information for improved object detection.
• ASFR-Net: Adversarial Alignment and Spatio-Frequency Refinement Network for Heterogeneous Remote Sensing Image Change Detection — ASFR-Net uses adversarial alignment and spatio-frequency refinement to improve heterogeneous remote sensing change detection.
• FS-DVS: A Frequency-Selective Dynamic Visual Sensing Paradigm for Enhancing Information Completeness — FS-DVS uses a learnable spatial filter to mimic biological vision, enhancing DVS information completeness and performance.
• PANet Paper Walkthrough: When Feature Pyramids Go Bottom-Up — PANet augments FPN with a bottom-up path to enrich deeper feature maps with spatial information, improving object detection accuracy.
• Dual-Integrated Low-Latency Single-Lens Infrared Computational Imaging for Object Detection — Integrating optical physics priors with deep learning for joint reconstruction-detection enables compact, low-latency infrared object detection.
• Revisiting Scene Graph Generation from the Perspective of Detector-Conditioned Reachability — Integrating detector-based and query-based SGG mechanisms via a dual-query design resolves their complementary prediction discrepancies.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.