tf-faster-rcnn

tf-faster-rcnn copied to clipboard

iter: 47220 / 70000, total loss: 0.354476

rpn_loss_cls: 0.000126 rpn_loss_box: 0.000068 loss_cls: 0.003523 loss_box: 0.000122 lr: 0.001000 speed: 0.276s / iter iter: 47240 / 70000, total loss: 0.355023 rpn_loss_cls: 0.000094 rpn_loss_box: 0.000046 loss_cls: 0.001107 loss_box: 0.003153 lr: 0.001000

speed: 0.276s / iter iter: 57020 / 70000, total loss: 0.353910

rpn_loss_cls: 0.000007 rpn_loss_box: 0.000003 loss_cls: 0.005448 loss_box: 0.000084 lr: 0.000100 speed: 0.276s / iter iter: 57040 / 70000, total loss: 0.349502 rpn_loss_cls: 0.000060 rpn_loss_box: 0.000010 loss_cls: 0.000791 loss_box: 0.000273 lr: 0.000100 speed: 0.276s / iter

iter: 69960 / 70000, total loss: 0.347819

rpn_loss_cls: 0.000047 rpn_loss_box: 0.000001 loss_cls: 0.000157 loss_box: 0.000086 lr: 0.000100 speed: 0.275s / iter iter: 69980 / 70000, total loss: 0.347693 rpn_loss_cls: 0.000048 rpn_loss_box: 0.000004 loss_cls: 0.000082 loss_box: 0.000033 lr: 0.000100 speed: 0.275s / iter iter: 70000 / 70000, total loss: 0.348334 rpn_loss_cls: 0.000065 rpn_loss_box: 0.000004 loss_cls: 0.000523 loss_box: 0.000217 lr: 0.000100 speed: 0.275s / iter

at first steps,loss is 0.35 and always 0.35,0.34 in 70000steps, why????

lr is too big??

loss_cls: 0.003523 loss_box: 0.000122 is small enough or overfitting

but at first 20 steps，the loss is 0.35 too, it does not change ............................

small enough .... ok try bigger..

I have the same problem ,and the loss_cls and loss_box are the 0 .why??

Is there anyone can tell me the reason why the loss_cls and loss_box are zero all the time when i trained???