Combining Transformers and CNNs for Efficient Object Detection in High-Resolution Satellite Imagery

TL;DR

This paper introduces GLOD, a transformer-based architecture optimized for object detection in high-resolution satellite images, achieving state-of-the-art performance through novel fusion and upsampling techniques.

Contribution

The paper proposes a transformer-first model with innovative fusion and upsampling modules, surpassing existing methods in satellite imagery object detection.

Findings

Achieves 32.95% mAP on xView dataset

Outperforms SOTA methods by 11.46%

Introduces asymmetric fusion with CBAM attention

Abstract

We present GLOD, a transformer-first architecture for object detection in high-resolution satellite imagery. GLOD replaces CNN backbones with a Swin Transformer for end-to-end feature extraction, combined with novel UpConvMixer blocks for robust upsampling and Fusion Blocks for multi-scale feature integration. Our approach achieves 32.95\% on xView, outperforming SOTA methods by 11.46\%. Key innovations include asymmetric fusion with CBAM attention and a multi-path head design capturing objects across scales. The architecture is optimized for satellite imagery challenges, leveraging spatial priors while maintaining computational efficiency.