trainloader issues

[CandiceUIC]

Hey I saw you updated your work recently, that is so nice of you.

I got a question about trainloader here. As I posted before, when I ran your code on my own computer, the loss function is increasing. The problem may be the trainloader. I found that in each epoch, the train loader only load about 480 images which is strange. The number of iterations in each epoch is only 6. I was trying to fix it by myself but it looks a little bit difficult to fully understand your code. Do you have any suggestions for me to fix this problem?

Cheers
Candice

[bnu-wangxun]

For every epoch, we do not use all the samples. We just randomly choose K (num_instances in the code) samples from each classes. So only 6 iterations for every epoch.

For the problem of loss function, I need to see your log to know the problem.

Ps: pytorch version is required to be 0.2.30

[CandiceUIC]

Thanks for your reply.
The version of pytorch I am using is 0.2.30. If you say that 6 iterations is correct, I would say that the codes are running well on my computer.
I am just a little bit curious why each epoch does not go through all the training set. That is a little bit weird to me.

[bnu-wangxun]

1. For easy Implementation, and the performance is good.
2. PK (P classes , K samples/class) minibatch. Since the number of samples in every class is different, it is hard to go through all the train set.

[bnu-wangxun]

@CandiceUIC Have you reached the performance as me?

[CandiceUIC]

@bnulihaixia For the Binomial Loss, I did reached a performance like you posted. But for other losses, for example triplet, I did not. I think that might be the reason of parameter choosing. I am now working on it.

Anyway, thank you so much for posting your work!

cheers