RGBS50
收藏资源简介:
RGBS50是首个针对水下RGB-Sonar(RGB-S)跟踪任务的基准数据集,由研究团队创建,包含50个时间对齐的水下视觉和声纳视频序列,总计超过87000个高质量标注。该数据集收集于深水池中,通过手动标注确保了目标边界框的准确性。RGBS50数据集旨在解决水下目标跟踪中由于光线散射和颜色退化导致的视觉限制,以及声纳图像中目标语义信息不足的问题。通过模拟训练方法SRST,数据集能够帮助模型学习RGB-S数据的语义结构,适用于水下多模态跟踪技术的发展,特别是在处理RGB和声纳图像间的空间错位问题上。
RGBS50 is the first benchmark dataset tailored for the underwater RGB-Sonar (RGB-S) tracking task, developed by a research team. It comprises 50 temporally aligned underwater visual and sonar video sequences, with a total of over 87,000 high-quality annotations. This dataset was collected in a deep underwater test tank, and the accuracy of target bounding boxes is guaranteed through manual annotation. The RGBS50 dataset is designed to address two core challenges in underwater target tracking: the visual limitations caused by light scattering and color degradation, and the insufficient semantic information of targets in sonar images. Leveraging the simulated training method SRST, this dataset enables models to learn the semantic structure of RGB-S data, and supports the development of underwater multimodal tracking technologies, particularly in resolving the spatial misalignment between RGB and sonar images.

- 1RGB-Sonar Tracking Benchmark and Spatial Cross-Attention Transformer Tracker未提及 · 2024年



