Dataset containing smoking and not-smoking images (smoker vs non-smoker)

The dataset contains a total of 2400 raw images, where 1200 images are of smoking (smokers) category and remaining 1200 images belong to no-smoking (non-smokers) category. The dataset is curated by scanning through various search engines by entering multiple keywords that include cigarette smoking, smoker, person, coughing, taking inhaler, person on the phone, drinking water etc. We tried to consider versatile images in both classes for creating a certain degree of inter-class confusion in order to better train the model. For instance, smoking category consists of images of smokers from multiple angles and various gestures. Moreover, the images in not-smoking category contains images of non-smokers with slightly similar gestures as that of smoking images such as people drinking water, using inhaler, holding the mobile phone, biting nails etc. The dataset can be used by the prospective researchers to propose machine learning algorithms for automated detection and screening of smoker towards ensuring the green environment and performing surveillance in smart cities.

本数据集共计包含2400张原始图像，其中1200张属于吸烟（smokers）类别，剩余1200张归属无烟（non-smokers）类别。本数据集通过输入多组关键词检索多家搜索引擎完成甄选构建，所涉关键词包括吸烟、吸烟者、人物、咳嗽、使用吸入器、打电话、饮水等。为优化模型训练效果，我们在两类样本中均选取了具备多样性的图像，以制造一定程度的类间混淆度。例如，吸烟类别涵盖了多视角、多姿态的吸烟者图像；而无烟类别则包含了与吸烟图像姿态略有相似的非吸烟者图像，如饮水者、使用吸入器者、持手机者、咬指甲者等。本数据集可供相关研究者开发机器学习算法，用于吸烟者的自动检测与筛查，助力绿色环境构建及智慧城市监控工作开展。