I am very excited about this new feature.
We have revamped similarity search, both in the back end and in the front end. Labelbox now supports high dimensionality embeddings, and the whole similarity search back end is faster and allows for instant searches.
On the front end, we have one-click similarity search, which allows you to browse your datasets very quickly by similarity. It is a great way to “learn” a dataset. Have you ever had to work with a big set of images that you haven’t seen before, and without knowing what’s inside? Similarity search is a great way to quickly browse it, gaining intuitions about what images are there.
More details in the blog post:
The new embeddings back end will also allow for some exciting new features, like clustering, better projection, and leveraging CLIP text queries. We are working on those and more right now, so the goodies will keep coming!
love me some 1-click similarity searchies.