Can you upload the loss of adaptation training here for comparison. The loss of loss_rpn_cls_s_fake, loss_rcnn_cls_s_fake and loss_rcnn_box_s_fake seem not converged in my training log, they are fluctuating.
What can be wrong in the training procedure?
Can you upload the loss of adaptation training here for comparison. The loss of loss_rpn_cls_s_fake, loss_rcnn_cls_s_fake and loss_rcnn_box_s_fake seem not converged in my training log, they are fluctuating. What can be wrong in the training procedure?