CMOTB
收藏资源简介:
CMOTB数据集是由安徽大学开发的一个大型跨模态目标跟踪视频数据集,包含654个序列,总计超过481,000帧,平均视频长度超过735帧。该数据集通过手持摄像机在多种场景和背景复杂度下捕捉,特别考虑了光照强度变化导致的模态切换,模拟了真实世界中的监控、智能交通和自动驾驶系统等应用场景。数据集的创建旨在解决传统RGB图像序列在低光条件下目标跟踪无效的问题,通过引入近红外(NIR)成像,克服了单一成像源的局限性。CMOTB数据集不仅用于训练深度跟踪器,还用于评估不同跟踪算法的性能,为跨模态目标跟踪的研究提供了重要的基准和资源。
The CMOTB dataset is a large-scale cross-modal object tracking video dataset developed by Anhui University. It contains 654 sequences with a total of over 481,000 frames, and the average video length exceeds 735 frames. Captured via handheld cameras across diverse scenarios and varying background complexity levels, the dataset specially considers modality shifts induced by illumination intensity changes, simulating real-world application scenarios such as surveillance, intelligent transportation, and autonomous driving systems. Developed to solve the problem that traditional RGB image sequences fail at object tracking under low-light conditions, the dataset overcomes the limitations of single imaging sources by introducing near-infrared (NIR) imaging. The CMOTB dataset can be used not only for training deep trackers but also for evaluating the performance of different tracking algorithms, providing an important benchmark and resource for cross-modal object tracking research.

- 1Cross-Modal Object Tracking: Modality-Aware Representations and A Unified Benchmark安徽大学 · 2021年



