Nima Tajbakhsh

I had a similar issue after upgrading to tf1.12. In my case, the error was thrown at the following line: tf.assert_equal(tf.shape(x), tf.shape(y))

I'm an engineer from Nvidia. I'm not able to reproduce this issue on my end. Basically, I built the image by `bash scripts/docker/build.sh` and then was able to start training...

I further tried to reproduce the issue you had: If I run the script with sbatch => it works; if I run it on an interactive node => it runs...

Thanks for the prompt reply! Theoretically, both operators should be +=, right? I mean the dice metric is class agnostic, so I assume the operator should be the same for...

Thanks. So, -= operator is acting as multiplying the gradient with -1, which should have been done for both bottom[btm].diff[i, 0, :] and bottom[btm].diff[i, 1, :], but you have intentionally...

All your explanations started to make sense after I re-read the paper and did the math by myself. Thanks!

@sagarhukkire Thanks for your reply. I read the closed thread but didn't find a link to the paper. Could you please point me to the reference?

Thanks! On Wed, May 3, 2017 at 2:41 PM, Sagar Hukkire wrote: > @ntajbakhsh > > https://arxiv.org/pdf/1701.03056.pdf > > — > You are receiving this because you were mentioned. >...