Fig 3. Counting performance of PrimSeq.

Shown are the mean ± standard deviation of the ground truth and predicted counts for patients in the test set; each dot represents a single subject. (A) Counts per primitive class. Activities were combined. The pipeline generated counts that were similar to true counts for each primitive class (mean percent of true counts: reach, 92.2%; reposition, 86.6%; transport, 99.5%; stabilization, 91.1%; and idle, 93.3%). (B) Counts per activity. Primitive classes were combined. The pipeline generated counts that were similar to true counts for each activity (mean percent of true counts: shelf task, 109.1%; tabletop task, 102.5%; feeding, 97.6%; drinking, 93.7%; combing hair, 89.8%; donning glasses, 85.2%; applying deodorant, 81.1%; washing face, 79.1%; and brushing teeth, 81.4%).