Hongjie Li

Hi @CharlesShang , I have the same question with @santisy In line 61 resnet50_test.py, the losses of pyramid_network were calculated but never use `**outputs = pyramid_network.build_losses(pyramid, outputs, gt_boxes, `gt_masks,` num_classes=81,...

@AaronYALai , Yes, the losses will be registered to tf.GraphKeys.LOSSES automatically, thanks for making this clear.

Hi @maxenceliu , How long it takes per iteration?

Hi @maxenceliu , has the total_loss explosion problem been solved with the newest version?