Build a Simplified Image–Text Retrieval Training and Evaluation Pipeline

Problem Build a simplified multimodal (image + text) retrieval pipeline that can be trained and evaluated, similar to how eBay matches product images to user text queries. What you are given A product dataset where each product has exactly one image. Title or description text for every product. Query texts, each linked to one or more positive product IDs. A pre-trained image encoder: resnet50. A pre-trained text encoder: sentence-transformers/all-MiniLM-L6-v2. What to…