search

wvangansbeke / Unsupervised-Classification

SCAN: Learning to Classify Images without Labels, incl. SimCLR. [ECCV 2020]
https://arxiv.org/abs/2005.12320
Other
1.37k stars 267 forks source link

embedding space #53

Closed smallelephant9516 closed 3 years ago

smallelephant9516 commented 3 years ago

It looks like an amazing work. However, I feel like I need some help. I have an unlabeled dataset and I want to get a general understanding of my dataset by looking at the higher level of representation on each image in the embedding space. Where I could check the embedding of each image to visualize the cluster? Thanks for your help.

wvangansbeke commented 3 years ago

Hi @smallelephant9516,

Thank for your interest and trying out the code.

You could "overcluster" your dataset and visualize the prototype of each cluster to get an idea what each cluster represents. I would also take a look at the nearest neighbors.

Have a look at eval.py to visualize the prototypes. I believe this will be helpful.

wvangansbeke commented 3 years ago

Hi @smallelephant9516,

If you have any more issues, let us know. Closing this for now.