While reviewing the first private release of WPSOLR’s demos with Weaviate, for the first time with realistic products, I realized that a vector search without image search made no sense.
It was not obvious until there, as I mostly played with dummy products.
But it is now clear that #retail search is deeply flawed if image content are not searched too.
WPSOLR is already able to call some entity extraction APis like Google Vision https://www.wpsolr.com/guide/configuration-step-by-step-schematic/activate-extensions/ai-image-and-ocr-apis-add-on/google-vision/, or Amazon Rekognition https://www.wpsolr.com/guide/configuration-step-by-step-schematic/activate-extensions/ai-image-and-ocr-apis-add-on/google-vision/, for tagging data during indexing.
But CLIP models are much more powerful https://www.wpsolr.com/guide/configuration-step-by-step-schematic/activate-extensions/ai-image-and-ocr-apis-add-on/google-vision/ and can search both descriptions, attributes, and images.
#woocommerce #weaviate #clipsearch #transformers #ml #retail #search