The evaluation accuracy (%) for bounding box generation and instance segmentation.

The upper part of the table is related to object detection (bounding boxes), and the lower part is related to instance segmentation. We used DCAN, CosineEmbedding, and Mask R-CNN for the other ANN architectures. When the entry is not applicable, dash (-) is used.表格上半部分对应基于边界框（bounding boxes）的目标检测（object detection）任务，下半部分则涉及实例分割（instance segmentation）任务。针对其余人工神经网络（Artificial Neural Network, ANN）架构，我们选用了DCAN、CosineEmbedding与Mask R-CNN。当条目不适用时，以短横线（-）标注。