Do you have a plan to release your train code?

[IrvingShu]

No description provided.

[fereenwong]

Waiting for train code +1 @almazan

[almazan]

Hi @IrvingShu and @fereenwong,

Unfortunately, we don't have plans for releasing the training code at the moment.

However, I just added to the repository the AP loss function that is described in the paper and that we used to train our models, which can be used for reference. You can find it here.

I hope that helps :)

Cheers!
Jon

[yan796]

hi, thanks for the code :). I was wondering if label should not be NxN ? Can you please help understand what label is from your equation (12) ? Is it Yi ?

[lyl472324464]

Hi @IrvingShu and @fereenwong,

Unfortunately, we don't have plans for releasing the training code at the moment.

However, I just added to the repository the AP loss function that is described in the paper and that we used to train our models, which can be used for reference. You can find it here.

I hope that helps :)

Cheers!
Jon

Why do we need loss function during reference? :o

[mrgransky]

@lyl472324464 for training process

[lyl472324464]

@lyl472324464 for training process

Got it.Thank you

[liyongqing-1101]

@yan796 @almazan Could you please simply explain why the shape of the label is N * M, why isn't N by N?

[Emalude]

@yan796 @almazan Could you please simply explain why the shape of the label is N * M, why isn't N by N?

I actually have a similar question. If the descriptors are N * M (images * embed_dimension) shouldn't the labels be just N? Maybe I am confusing embedding and descriptors?