SARDet-100K: Towards Open-Source Benchmark and ToolKit for Large-Scale SAR Object Detection

Synthetic Aperture Radar (SAR) object detection has gained significantattention recently due to its irreplaceable all-weather imaging capabilities.However, this research field suffers from both limited public datasets (mostlycomprising <2K images with only mono-category objects) and inaccessible sourcecode. To tackle these challenges, we establish a new benchmark dataset and anopen-source method for large-scale SAR object detection. Our dataset,SARDet-100K, is a result of intense surveying, collecting, and standardizing 10existing SAR detection datasets, providing a large-scale and diverse datasetfor research purposes. To the best of our knowledge, SARDet-100K is the firstCOCO-level large-scale multi-class SAR object detection dataset ever created.With this high-quality dataset, we conducted comprehensive experiments anduncovered a crucial challenge in SAR object detection: the substantialdisparities between the pretraining on RGB datasets and finetuning on SARdatasets in terms of both data domain and model structure. To bridge thesegaps, we propose a novel Multi-Stage with Filter Augmentation (MSFA)pretraining framework that tackles the problems from the perspective of datainput, domain transition, and model migration. The proposed MSFA methodsignificantly enhances the performance of SAR object detection models whiledemonstrating exceptional generalizability and flexibility across diversemodels. This work aims to pave the way for further advancements in SAR objectdetection. The dataset and code is available athttps://github.com/zcablii/SARDet_100K.