Action recognition and anticipation of regular and just in-time procedures The task of action recognition and anticipation in medical procedures challenges participants to simultaneously recognize current actions and predict next actions on egocentric medical procedure videos. This dual-task formulation closely mirrors real-world scenarios and needs in emergency medicine, battlefield medicine, and disaster response, where accurate and timely interpretation of medical actions is crucial for effective intervention. A key challenge in this task is the diverse and uncontrolled nature of medical environments, where procedures are often unscripted and highly variable. Numerous action classification benchmarks exist in the field, such as UCF101, HMDB51, Kinetics, Something-Something, and AVA. However, the Trauma THOMPSON dataset stands out as the first egocentric video dataset specifically designed for medical procedure action recognition and anticipation, bringing unique value to the field.

Emergency procedure hand tracking Hand tracking is a challenging but significant task. Such algorithm can assist care providers to identify proper tools in a more straightforward manner. The task involves precisely capturing hand movements, classifying hand types (left or right hand), and recognizing medical tools in real-time. The ability to perform these tasks effectively is integral to enhancing decision-making in critical situations, and the presence of gloves makes these tasks even harder. The challenge focuses on advancing algorithms that can achieve high levels of accuracy and precision in detecting and tracking hands, even under the unpredictable conditions often encountered in medical environments. The decent performance of the task will largely improve the efficiency and quality of the surgeries. Added to it, it ensures the safety of patients, especially in urgent emergencies when patients meet the threaten of death.

Emergency procedure tool detection Emergency procedure tool detection is a fundamental yet critically important task in the Trauma Thompson Challenge. This type of algorithm is designed to detect crucial medical instruments throughout the entire emergency response process. Assisted by hand tracking, the algorithm aids healthcare providers in swiftly identifying every medical device involved in an emergency scenario, thereby enabling precise emergency support measures. Therefore, accurate identification of each emergency tool is of utmost importance. Operating under various unpredictable medical conditions, recognizing medical instruments that may differ in color, shape, but share the same function remains highly challenging. An accurate and broadly applicable emergency tool detection algorithm serves as the cornerstone for implementing emergency procedures, ensuring that the patient's emergency care process is conducted rapidly, correctly, and effectively.

Realism assessment The realism assessment task requires participants to identify how realistic a simulated trauma procedure video appears from an egocentric point of view. Algorithms must be able to identify the realism of a video segment, using information about the procedure including steps, tools, and other important aspects of surgery. This would allow for the curation of higher quality training datasets, which would allow algorithms to learn the most accurate and realistic surgical processes, an essential step of improving accuracy in trauma care delivery assistance.

Visual Question Answering The visual question answering (VQA) task requires participants to answer various questions about egocentric images from surgical procedures. They must be able to identify the correct answers to any given question relating to tools, next steps, and other components of surgery. Although multiple VQA datasets exist (VQAv2, Coco, etc.), the Trauma THOMPSON VQA dataset is unique as it is the first to utilize egocentric trauma surgery data when prompting the participant for an answer. The algorithms implemented by participants must be innovative enough to analyze complex trauma surgery images and be able to generate an answer to any question related to that image. This aids in more accurate decision-making and less confusion among healthcare providers performing trauma care procedures.