Hi! Is there any plan to embed similarity search into Uploadcare? Currently to find visually similar images we use pytorch with ResNet-152 model. It doesn’t take that much code to do feature extraction and approximate neighbourhood search (we are using HNSW index), but overall infrastructure for doing that gets really bulky. So I though maybe Uploadcare is doing something for that already as the topic seems to be really trending now.
Yes, we’re thinking of a way to add “similar search” to the service.
Can’t give any estimations at this point though.
We’ve added perceptual hash to file information, so now it’s easy to find image duplicates and similar images by quickly comparing their perceptual hash values. The hash is available through the