zhang6ting/fisheye8k

FishEye8K是一个用于鱼眼相机物体检测的基准和数据集，包含8,000个样本，分为训练集5,288个样本和测试集2,712个样本。数据集由18个鱼眼相机拍摄的22个视频中的8,000张图像组成，分辨率为1080×1080和1280×1280，用于交通监控。数据集包含157K个边界框，覆盖五个类别（行人、自行车、汽车、公交车和卡车）。数据集由Munkhjargal Gochoo等人策划，Harpreet Sahota共享，使用英语，许可证为Creative Commons Attribution-NonCommercial-ShareAlike 4.0。数据集来源包括GitHub仓库、网站和相关论文。数据集结构遵循特定文件名格式，每个样本有两个专用元数据字段存储位置和时间信息。数据集创建过程包括源数据采集、注释规则和注释过程，以及数据匿名化处理以保护隐私。

FishEye8K is a benchmark and dataset for fisheye camera object detection, comprising 8,000 samples with 5,288 in the train set and 2,712 in the test set. The dataset consists of 8,000 images from 22 videos recorded by 18 fisheye cameras for traffic monitoring in Hsinchu, Taiwan, at resolutions of 1080×1080 and 1280×1280. It includes 157K bounding boxes across five classes (Pedestrian, Bike, Car, Bus, and Truck). The dataset is curated by Munkhjargal Gochoo et al., shared by Harpreet Sahota, in English, under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 license. Dataset sources include a GitHub repository, a website, and a related paper. The dataset structure follows a specific filename format, with each sample having two dedicated metadata fields for location and time of day. The dataset creation process involves source data collection, annotation rules and process, and data anonymization for privacy protection.