In my experiments (a different setup but using discriminative loss), I am finding the loss computation to be very time-consuming. I suspect the lag is coming from the tf.unsorted_segment_sum. Have you noticed such latency issues in your experiments? Is there perhaps a way to speed up/optimize the loss computation?
Hi, This is Parth Dave.
I am trying to run this repository on my machine. Can you please give me the requirements in order to run this repo successfully?
In my experiments (a different setup but using discriminative loss), I am finding the loss computation to be very time-consuming. I suspect the lag is coming from the
tf.unsorted_segment_sum
. Have you noticed such latency issues in your experiments? Is there perhaps a way to speed up/optimize the loss computation?