In this line the normalisation is done by applying the reduce_mean. Usually I would expect to see the sum, which would be L2 normalisation and would keep the vector at unit variance.
There is a normalisation strategy mentioned in the paper, however it does not match what has been implemented.
https://github.com/zsyzzsoft/co-mod-gan/blob/c8b9ffe30c950dfdfb9e86652a27b6e72304b3d5/training/co_mod_gan.py#L285C1-L286C1
In this line the normalisation is done by applying the
reduce_mean
. Usually I would expect to see the sum, which would be L2 normalisation and would keep the vector at unit variance.There is a normalisation strategy mentioned in the paper, however it does not match what has been implemented.
Would it be possible to get some clarification?
Thanks