thanhhoangnvbg/fire-vn-yolo11seg-v1
收藏资源简介：
Fire VN YOLO11-Seg v1 是一个准备好的火灾和烟雾分割数据集，专注于越南视觉环境。它旨在训练和评估YOLO11分割模型，用于早期火灾/烟雾检测，场景包括城市小巷、居民区、远距离火灾/烟雾事件、困难负样本视觉干扰物和正常环境场景。数据集格式为YOLO分割，包含smoke（烟雾）和fire（火灾）两个类别，总图像数为15,036张，总标注数为30,054个。数据集分为五个源组：清晰火灾/烟雾图像、越南特定小巷/城市环境、困难负样本、远距离或小物体火灾/烟雾、正常环境场景。划分包括训练集（13,246张图像，25,486个标注）、验证集（957张图像，2,214个标注）和测试集（833张图像，2,354个标注）。数据集通过从Roboflow导出、独立处理源组、确定性平衡分割、训练集切片和转换为YOLO分割格式准备。目录结构包括data.yaml、图像、标签、COCO格式和元数据文件夹。数据集用于火灾/烟雾检测系统研发、YOLO11分割训练、内部模型变体基准比较以及越南环境早期预警模型的微调改进，但不应作为生产部署安全的唯一证据，仍需实际摄像头验证和操作误报测试。
Fire VN YOLO11-Seg v1 is a prepared fire and smoke segmentation dataset focused on Vietnamese visual contexts. It is intended for training and evaluating YOLO11 segmentation models for early fire/smoke detection in scenes such as urban alleys, residential areas, distant fire/smoke events, hard negative visual distractors, and normal ambient scenes. The dataset format is YOLO segmentation, with classes smoke and fire, total images after training-only slicing of 15,036, and total annotations of 30,054. The dataset is prepared from five source groups: clear fire/smoke images, Vietnam-specific alley/urban context, hard negatives that may look like fire/smoke, distant or small fire/smoke objects, and normal ambient scenes. Splits include train (13,246 images, 25,486 annotations), valid (957 images, 2,214 annotations), and test (833 images, 2,354 annotations). The preparation method involves reading source groups from Roboflow exports, creating a deterministic balanced split, applying slicing only to the training split, and converting to YOLO segmentation format. Directory structure includes data.yaml, images, labels, coco, and metadata folders. The dataset is intended for research and development of fire/smoke detection systems, YOLO11 segmentation training, benchmark comparisons between internal model variants, and fine-tuning early warning models for Vietnamese environments, but should not be used as the only evidence for production deployment safety, requiring real-world camera validation and operational false-alarm testing.
