Do you load the entire training dataset in the memory bank?

[trisct]

Thanks for this work.

There is one thing that is a bit unclear. Exactly what is stored in the memory bank? Do you store all the training images in there, or only a subset of images? It seems that you are loading the whole imagenet into the bank, which takes a lot of memory space. Since I am trying to reproduce on my PC, this is impossible. Is there a commendation on the lower bound of the dataset size that does not severely affect the result?

[chengxuz]

The memory bank stores the embeddings of all the training images.

[chengxuz]

To be clear, the memory bank is in the shape of 1.2M * 128, where 128 is the dimension of the embedding, and the 1.2M is the size of ImageNet.