Figure 4
The “object-in-object” phenomenon. (A) The instance segmentation is good for discrimination of different people overlapping with each other. (This sample image was the prediction of Mask R-CNN on COCO dataset20.) (B) However, one wound may contain multiple textures. (C) In the training set, we could identify different textures in these two solitary wounds. (D) Mask-RCNN inaccurately performed segmentation, identifying a second smaller “wound” with different texture (blue) within the bigger wound (red). However, in actuality, there was only one wound.
