Figure 2.
Overview of the instance segmentation and tracking frameworks. (a) The instance segmentation framework was trained with three types of training datasets: the bilateral axillo-breast approach (BABA) training model, patients, and a public database. (b) The tracking framework, consisting of a tracker and a sequence of re-identification algorithms. Spatial-temporal re-identification (ST-ReID) was trained with bounding boxes of all types of surgical instruments. Bag of visual words re-identification (BOVW-ReID) was applied after ST-ReID.