@alg@alghttps://teletype.in/atom/alg2026-05-02T09:34:28.269Zalg:imageclusterImage clustering2020-05-08T11:24:34.341Z2020-05-08T11:24:34.341Z<img src="https://teletype.in/files/82/a2/82a2bf21-a873-490b-bbcf-ee8f2b99bdc4.png">I've used https://github.com/elcorto/imagecluster. It computes neural feature vector via VGG16. Uses around 100G RAM for 20k photos. Works OK on cpu, if you have like 16 of them.
<figure class="m_original">
<img src="https://teletype.in/files/82/a2/82a2bf21-a873-490b-bbcf-ee8f2b99bdc4.png" width="815" />
<figcaption>Thanks to all the people on social networks, who put much effort organized images into nice datasets on every imaginable topic</figcaption>
</figure>
<p>I've used <a href="https://github.com/elcorto/imagecluster" target="_blank">https://github.com/elcorto/imagecluster.</a> It computes neural feature vector via VGG16. Uses around 100G RAM for 20k photos. Works OK on cpu, if you have like 16 of them.</p>
<p>For clustering, sqeuclidean metric worked best for me. </p>
<p>There's also <a href="https://github.com/facebookresearch/deepcluster" target="_blank">https://github.com/facebookresearch/deepcluster,</a> but all it did for me is mess with CUDA and crash.</p>
<p>Bonus picture from dataset</p>
<figure class="m_original">
<img src="https://teletype.in/files/74/4f/744f13f8-5568-48aa-9ff7-8b77420d5004.png" width="1024" />
</figure>