K-means for SpCL

[yjh576]

Have you tried K-means based clustering to generate pseudo labels for SpCL? I tried it based on your open source code. However, using K-means based pseudo labels for SpCL, we got poor results. For example, mAP are 20.9% and 36.2%,respectively. When the pseudo labels of MMT are generated by K-means, its results is comparable with that of DBSCAN. I can't understand and explain this. Could you give some suggestions? Thank you in advance.