COSOS-1k: A Benchmark Dataset and Occlusion-aware Uncertainty Learning for Multi-view Video Object Detection. — SciRadar