"a simulated dataset tailored for dim and small space targets under space-based observation."

"Since it is challenging to obtain real-world image sequences of dim and small space targets under space-based observation, we employ the aforementioned simulation framework to generate a synthetic dataset, SIMDataset, comprising 57 groups of image sequences for training and testing. This dataset simulates a space-based detection scenario, where a satellite located approximately 420,000 km from Earth observes high-orbit targets. SIMDataset is specifically designed for the task of dim and small target detection in space-based imagery. It includes various challenging conditions such as stray light interference, non-uniform illumination, and no interfering illumination. The SNR of the targets ranges from 0.8 \u223c 2.0, and targets move by 1 \u223c 2 pixels per frame on average. Each image has a resolution of 500 \u00d7 500 pixels, and target sizes range from 1 \u00d7 1 to 3 \u00d7 3 pixels. Each sequence contains 1 to 7 moving targets. The dataset is annotated in the COCO format, which includes fields such as images, annotations, and categories. It is divided into 39 sequences containing 3,900 images for training and 18 sequences with 1,800 images for testing."