I use a convolutional neural network based feature extraction and encoded results. This shows 5,623 photos cluster the similar artists based on 25,088 dimensions. [1]

The mouse in this link [2] shows two artists, Banksy and Blek le Rat – who has similar styles.

Here is the code [3] for the encoding process (again all by Dhruv from @Cometml [4])

I can share the dataset of for anyone who is interested. There are 15 artists and ~5300 photos. Shoot me a message on Twitter [5].

The encoded features (using VGG16) are in a hdf5 file that I can also share for anyone interested.

The clustering is done using the Annoy (Approximate Nearest Neighbors OY) [6] from @SpotifyEng – I tried to use FAISS [7], but struggled with the instructions. Apparently its significantly faster, but the documentation wasn’t great.

The visual gifs are generated using Tensorflow.js in the Embeddings Projector – a very cool project that runs off a basic HTML file in your local . More info here:

[^1] https://twitter.com/rememberlenny/status/1038992069094780928

[^2] https://twitter.com/rememberlenny/status/1038994489896042497

[^3] https://github.com/rememberlenny/graffiti-net

[^4] https://www.comet.ml/lenny

[^5] https://twitter.com/rememberlenny

[^6] https://github.com/spotify/annoy

[^7] https://github.com/facebookresearch/faiss

Source link
thanks you RSS link
( https://www.reddit.com/r/MachineLearning/comments/9ejdq7/p___images_with_machine_learning/)


Please enter your comment!
Please enter your name here