Behaviour agreement among the two graders (top section) and quality of the system compared to the two graders (middle and bottom sections) on Dataset A (higher values are better ).

The agreement is computed on a frame-by-frame basis. Each value represents the percentage of frames with class agreement. The different columns show the results for different types of behaviours: ★accuracy on all behaviours considered separately; †accuracy on behaviours grouped into social and non-social meta-classes; ‡precision on the social behaviours; °precision on the non-social behaviours. Accuracy = (TP+TN)/(TP+TN+FN+FP) and Precision = TP/(TP+FP) where TP = True Positive, FP = False Positive, FN = False Negative and TN = True Negative. The average agreement grader/system is comparable to the average grader/grader agreement.