the segm mAp result is zero

[YangPanHZAU]

I used my own dataset to train the instance segmentation task on K-Net and found that the seg mAP was always 0, and the number of training epochs still did not change，my num_classes=1.