search

jwyang / JULE.torch

Torch code for our CVPR 2016 paper "Joint Unsupervised LEarning of Deep Representations and Image Clusters"
MIT License
286 stars 82 forks source link

Number of minimum cluster items #9

Open bnascimento opened 7 years ago

bnascimento commented 7 years ago

Is there anyway we can control the minimum number of cluster items? E.g. Would like to avoid having a cluster with 2 items.

Thank you Bruno

jwyang commented 7 years ago

Hi, Bruno,

The recent version cannot control the number of samples in a single cluster directly. For in-directional methods, you might change K_s to get more neighbors into account. Also, I used merge_cluster to merge some small clusters in the code. It should get around your mentioned case. Did it not work in your case?

bnascimento commented 7 years ago

K_s didn't help much, I still get fragmented clusters with just 2 images. Tested ranges from 5 to 100