GUS-FPN Evaluation Protocol and MMDetection-Compatible Toolkit
Abstract
This repository provides a lightweight, MMDetection-compatible release accompanying the paper “GUS-FPN: A Spectral-Gating, Uncertainty-Guided, and Spatial-Dependency Co-Enhanced FPN for Dense Small Object Detection/Segmentation in Aerial Imagery”. Contents:(1) An evaluation protocol for dense small-object analysis, including Dense/NonDense subset construction and two evaluation modes (COND and BG) for fair assessment under background clutter.(2) Ready-to-run testing/evaluation scripts compatible with MMDetection/MMEngine.(3) Reproducible model configuration files used in our experiments. What this release does NOT include:- The iSAID dataset (users must obtain it from the official source and follow its license/terms).- Any original MMDetection source files (this release depends on MMDetection as an external library and respects the upstream license). Usage:After installing MMDetection, users can directly run the provided scripts and configs to reproduce the reported evaluation results. See README for installation notes, dataset path conventions, and example commands. License:This software is released under the Apache-2.0 license. Third-party dependencies and attribution notes are documented in NOTICE/THIRD_PARTY_NOTICES. For questions or issues, please contact the corresponding author of the paper.
Fields of Study & Subjects
Advanced Neural Network Applications
Remote-Sensing Image Classification
Visual Attention and Saliency Detection
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 30 January 2026