Retrieval using images
Hi,
I'm interested in the possibility of using images as both documents to embed and as query against these stored documents, is this a possibility with this model?
I note in the 'modeling_jina_embeddings_v4.py' file the public interface for 'encode_text' there is a the 'prompt_name: Type of text being encoded ('query' or 'passage')' which applies to the 'retrieval' task, and there does not appear to be this for the 'encode_image' method. Is the above case not possible or is it that the retrieval task is does not differentiate between images as passages or as queries?
Thanks for any help.
We only trained the model for text-to-image search (for pictures and screenshots of documents). Therefore, we also did not consider different prompts for images. So it might not be ideal for image-to-image search. Nevertheless, you can still try this. I wouldn't use any specific prompts for queries and documents in this case.