Powering Semantic Similarity Search in Computer Vision with State of the Art Embeddings