similarity_search_by_image_with_relevance_score() — langchain Function Reference

Source Code

libs/partners/chroma/langchain_chroma/vectorstores.py lines 986–1024

    def similarity_search_by_image_with_relevance_score(
        self,
        uri: str,
        k: int = DEFAULT_K,
        filter: dict[str, str] | None = None,  # noqa: A002
        **kwargs: Any,
    ) -> list[tuple[Document, float]]:
        """Search for similar images based on the given image URI.

        Args:
            uri: URI of the image to search for.
            k: Number of results to return.
            filter: Filter by metadata.
            **kwargs: Additional arguments to pass to function.

        Returns:
            List of tuples containing documents similar to the query image and their
            similarity scores. 0th element in each tuple is a LangChain Document Object.
            The page content is b64 encoded img, metadata is default or defined by user.

        Raises:
            ValueError: If the embedding function does not support image embeddings.
        """
        if self._embedding_function is not None and hasattr(
            self._embedding_function, "embed_image"
        ):
            # Obtain image embedding
            # Assuming embed_image returns a single embedding
            image_embedding = self._embedding_function.embed_image(uris=[uri])

            # Perform similarity search based on the obtained embedding
            return self.similarity_search_by_vector_with_relevance_scores(
                embedding=image_embedding,
                k=k,
                filter=filter,
                **kwargs,
            )
        msg = "The embedding function must support image embedding."
        raise ValueError(msg)