Open JohnTigue opened 3 years ago
[John]():
During stand-up Jed made me realize that one thing we could do is use some commercial image labeler to generate, say, top-ten lists of words that could be machine generated labels for an image (in this use case, an image would be a work of TreasurePop art). Then we could take those words (not the images like I'm doing now) and cluster plot them. word2vec, or some NLP embedder or such. The goal would be for the ML to do a sort of "topic mapping" whereby similar words would be clustered. This tool could be used to find themes hidden in a collection of TreasurePop images.
https://umap-learn.readthedocs.io/en/latest/document_embedding.html