Evaluation Metrics

1 segm_mAP（Segmentation Mean Average Precision）

segm_mAP (Segmentation Mean Average Precision) is the core evaluation metric for the instance segmentation task, which is used to measure the comprehensive performance of a model in pixel-level segmentation and classification. It is obtained by calculating the average precision (AP) of the mask intersection-over-union (IoU) for each category at different thresholds and then taking the mean of the APs of all categories. This metric strictly evaluates the positioning accuracy and classification correctness of the model for target masks, with particular attention to the performance on different scales (small/medium/large targets). It serves as the standard evaluation basis for authoritative datasets such as COCO. A higher value indicates a stronger segmentation ability of the model.