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

Authors: Yuxuan Li, Xiang Li, Weijie Li, Qibin Hou, Li Liu, Ming-Ming Cheng, Jian Yang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental With this high-quality dataset, we conducted comprehensive experiments and uncovered a crucial challenge in SAR object detection: the substantial disparities between the pretraining on RGB datasets and finetuning on SAR datasets in terms of both data domain and model structure. 5 Experiments and Analysis
Researcher Affiliation Academia 1 PCA Lab, VCIP, CS, Nankai University 2NKIARI, Futian, Shenzhen 3Academy of Advanced Technology Research of Hunan
Pseudocode No The paper does not contain any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The dataset and code is available at https://github.com/zcablii/SARDet_100K. Establishment of a new benchmark in SAR object detection by releasing the datasets and code associated with our research.
Open Datasets Yes Our dataset, SARDet-100K, is a result of intense surveying, collecting, and standardizing 10 existing SAR detection datasets, providing a large-scale and diverse dataset for research purposes. The dataset and code is available at https://github.com/zcablii/SARDet_100K.
Dataset Splits Yes If the source dataset already provides predefined train, validation, and test splits, we adopt their split settings. Otherwise, we perform the splitting by ensuring a ratio of 8:1:1 for the train, validation, and test sets respectively.
Hardware Specification Yes We primarily conduct our experiments using the MMPretrain [11] and the MMDetection [8] frameworks, on 8 RTX-3090 GPUs (24G).
Software Dependencies No We primarily conduct our experiments using the MMPretrain [11] and the MMDetection [8] frameworks - no version numbers are provided for these frameworks.
Experiment Setup Yes For detailed information on the hyperparameters and training settings, please refer to Table S12. Table S12: Hyper-parameter of pretrain and finetune settings. Cls.: Classification, Det.: Detection, B.S.: Batch Size, L.R.: Learning Rate. Task / Model Dataset Optim. B.S. L.R Epochs