I hope all is well. I am currently working on similarity search using both text and images, and I was wondering, based on your opinion, if there was any better approach than extracting embedding vectors and perform a cosine similarity search.
Also, has anyone implemented DeViSE in v1.0 of the library?
There’s a DeViSe notebook available that does Image to Image and Text to Image (and reverse) available on the course repo https://github.com/fastai/course-v3/blob/master/nbs/dl2/devise.ipynb
I have tired it but the results weren’t quite as good as traditional Cosine Similarity search.