Labeled Images of a Simulated Phlebotomy Procedure on a Training Arm

Description This dataset contains a collection of 11,884 annotated images derived from a simulated blood drawing procedure performed on a medical training arm. The images were extracted from third-person videos and depict a sequence of standard medical actions, such as applying a tourniquet, disinfecting the skin, and inserting a syringe. The dataset is designed to facilitate computer vision tasks in the medical field, specifically those related to human-object interaction (HOI). The annotations follow the YOLOv8 (You Only Look Once) format, and each image includes bounding boxes corresponding to key objects involved in the procedure. The goal is to provide a resource for training and evaluating object detection models in simulated clinical environments, thereby supporting intelligent feedback systems for medical training. Purpose The primary objective of this dataset is to support the development of: - Detecting medical objects in clinical scenarios - Understanding scenes in training simulations - Recognizing human-object interactions during medical procedures - Assessing procedural skill and guiding learners using deep learning models This dataset is aimed at researchers and practitioners in medical AI, computer vision, and simulation-based training. Dataset Overview - Total images: 11,884 - Image resolution: Variable, with longer side 640 px (aspect ratio preserved; e.g., 640×480, 640×360). - Format: JPEG files with corresponding YOLOv8 `.txt` annotations - Annotation type: Bounding boxes in normalized YOLOv8 format Classes and Label Counts - Class 0: Syringe – 9,949 instances - Class 1: Elastic Band – 6,588 instances - Class 2: Disinfectant Wipe – 1,152 instances - Class 3: Training Arm – 28,370 instances - Class 4: Gloves – 17,168 instances Data Split - Training set: 70% - Validation set: 15% - Test set: 15% The split is provided as a suggestion and can be adjusted depending on experimental needs.