I've been training an implementation of Mask R-CNN and it was training very successfully on my CPU but I've just set up my GPU and it is giving some strange results when looking at my validation loss. It is as if the learning rate got turned up when I moved to using my GPU. Has anyone ever experienced this or know what could cause this difference? I didn't think there was any difference in the actual computation on CPU vs GPU, is there?

Training on CPU:

enter image description here

Training on GPU:

enter image description here

  • $\begingroup$ Have you trained it from scratch using GPU? $\endgroup$ Jan 8, 2019 at 22:34
  • $\begingroup$ This model is pre-trained using COCO which I would assume was done with a GPU but it wasn't done by me. $\endgroup$
    – clifgray
    Jan 9, 2019 at 2:08


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.