GeoDec-Net: A Geometry-Decomposed Network for Oriented Object Detection in Remote Sensing Images

This repository provides the official PyTorch implementation of the paper "GeoDec-Net: A Geometry Decomposed Oriented Bounding Boxes Object Detection Network for Remote Sensing Images". GeoDec-Net is a lightweight single-stage oriented object detector designed for remote sensing imagery, incorporating three core modules: the Geometry-Decomposed Block (GDBlock) for multi-dimensional geometric feature extraction, Geometry-Decomposed Pooling (GDPool) for anisotropic long-range context modeling of slender objects, and Geometry-Decomposed Fusion (GDFusion) for high-resolution directional feature refinement and background noise suppression. This archive contains the complete network architecture code, model configuration files, and scripts for training, validation, and inference latency benchmarking on the DOTA-v1.0 and HRSC2016 datasets.