wenguanwang
commented
2 years ago @RenLibo-aircas thanks for your interest. Basically, the equipartition constraint (see Eq. 8) in Sinkhorn-Knopp iteration can improve the diversity of the cluster centers. But in practice, case-by-case consideration is needed. For example, if the patterns of the semantic classes of your interest are simple or the number of training data is limited, there is no need to adopt a large number of prototypes.
Hi, I'm interested in your work. I wonder if there are k prototypes of some classes that become similar after training? For example, after visualization according to Figure 3 in the paper, it will be found that the activation area of each prototype is roughly the same. I found this problem while running your code. I suspect it's caused by some classes of my dataset that don't have meaningful parts.
Hope to receive your reply, thanks!