Upload M4-SAR.zip

Single-source remote sensing object detection, which relies solely on either optical or Synthetic Aperture Radar (SAR) images, often underperforms in real‐world scenarios. Although rich in texture and color, optical images degrade significantly under cloud cover, low resolution, or poor illumination. Conversely, SAR imagery offers all-weather imaging but suffers from blurred visual characteristics that limit its effectiveness. These limitations severely constrain detection accuracy and robustness in critical applications such as disaster monitoring and urban planning. Multi-source fusion of optical and SAR data can effectively address challenges posed by cloud cover and low visibility. However, the lack of a large-scale, standardized baseline has hindered progress in this field. To address this gap, we propose the first comprehensive benchmark dataset for optical-SAR fusion object detection, called Multi-resolution, Multi-polarization, Multi-scene, Multi-source SAR dataset (M4-SAR). The dataset contains 112,184 precisely aligned image pairs and nearly one million labeled instances in arbitrary orientations, covering six key categories such as bridges, ports, and airports.

M4-SAR.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ca1ae75ec190e815f84018c5f478571ef4ee1a11003f6565ebffe5de8812633
+size 20531528868