CigDet (Cigarette Detection) Dataset

Our dataset, named CigDet, is tailored to advance the field of cigarette detection. Leveraging the smoker detection dataset available on the Mendeley Data online repository which was our previous work, we specifically concentrated on images categorized under the Smoking class. In these images, the presence of cigarettes serves as the pivotal attribute for smoker detection. Each image in our dataset is meticulously chosen to depict various scenes containing cigarettes amidst diverse backgrounds and environmental conditions, effectively simulating real-world complexities encountered in cigarette detection tasks.

For precise annotations, we utilized the LabelImg tool, renowned for its efficacy in annotation tasks, adhering strictly to the YOLO format. Employing a streamlined annotation approach, we focused solely on delineating and annotating instances of cigarettes within the images, ensuring accuracy and relevance.

Our dataset comprises 557 images, each representing a valuable sample for the intended task. To facilitate robust model training and thorough evaluation, we divided the dataset into distinct subsets for training and testing, comprising 446 and 111 images, respectively. This partitioning strategy was devised to ensure equitable distribution of instances between training and testing sets, enabling comprehensive evaluation of the model's performance across diverse scenarios.