Fire360
收藏资源简介:
Fire360是一个用于评估在烟雾、低光和结构变形等恶劣环境下,消防场景中感知和推理能力的基准数据集。该数据集包含228个360度视频,这些视频来自专业的消防训练课程,并在多种条件下(如低光、热失真)进行了标注,包括动作片段、对象位置和退化元数据。Fire360支持五个任务:视觉问答、时间动作描述、对象定位、安全关键推理和转换对象检索。通过发布Fire360及其评估套件,旨在推进模型在不确定性条件下的感知、记忆、推理和行动能力。
Fire360 is a benchmark dataset for evaluating perception and reasoning capabilities in firefighting scenarios under harsh environments such as smoke, low-light conditions, and structural deformations. This dataset contains 228 360-degree videos sourced from professional firefighting training courses, which are annotated under various conditions including low-light and thermal distortion, covering action clips, object positions and degradation metadata. Fire360 supports five tasks: visual question answering, temporal action captioning, object localization, safety-critical reasoning and transformed object retrieval. By releasing Fire360 and its evaluation suite, the aim is to advance the development of models' perception, memory, reasoning and action capabilities under uncertain conditions.
Fire360 Dataset 概述
数据集内容
- pristine_objects_for_TOR: 34个文件
- extra_video_folders: 242个文件
- daytime_videos: 80个文件
- night_time_videos: 91个文件
- indoor_videos: 134个文件
- 样本视频文件:
- sample_9.MP4: 1.1 GB
- sample_0.MP4: 2.6 GB
- sample_10.MP4: 137.2 MB
- sample_1.MP4: 142.7 MB
- sample_2.MP4: 83.7 MB
- sample_3.MP4: 49.3 MB
- sample_4.MP4: 53.9 MB
- sample_5.MP4: 241.4 MB
- sample_6.MP4: 906.8 MB
- sample_7.MP4: 197.3 MB
- sample_8.MP4: 1.8 GB
- 文档文件:
- NeurIPS_2024.pdf: 97.5 KB
- splits.json: 5 KB
数据集属性
- 所有者: Aditi Tiwari
- 所属机构: University of Illinois
- 创建时间: 2025年5月9日 6:03 AM
- 修改时间: 2025年5月16日 8:13 PM
- 总大小: 484.7 GB

- 1Fire360: A Benchmark for Robust Perception and Episodic Memory in Degraded 360-Degree Firefighting Videos伊利诺伊大学香槟分校 · 2025年



