Multi-Modal and Multi-Domain Embedding Learning for Fashion Retrieval and Analysis