A question about computing Ci by k-mean

[MrYaoH]

Hi, chengxuz, I have a question. In LA part, the loss backpropogates before k-mean updation. Does this mean that the first data batch did not define Ci as mentioned in the paper？

[chengxuz]

At the begining of the training, which is before any loss computation, Ci would be computed using the saved memory bank.