I find the maskIoU branch does not play a role in inference, does it mean that i can pruning this branch after training? Therefore, the maskIoU only influences in back propagation?
The maskIoU branch is used in the testing phase, the predicted maskiou multiply with the classification score, which can improve the mAP by about 1.5 points.
I find the maskIoU branch does not play a role in inference, does it mean that i can pruning this branch after training? Therefore, the maskIoU only influences in back propagation?