Jina CLIP: Your CLIP Model Is Also Your Text Retriever

About

Contrastive Language-Image Pretraining (CLIP) is widely used to train models to align images and texts in a common embedding space by mapping them to fixed-sized vectors. These models are key to multimodal information retrieval and related tasks. However, CLIP models generally underperform in text-only tasks compared to specialized text models. This creates inefficiencies for information retrieval systems that keep separate embeddings and models for text-only and multimodal tasks. We propose a novel, multi-task contrastive training method to address this issue, which we use to train the jina-clip-v1 model to achieve the state-of-the-art performance on both text-image and text-text retrieval tasks.

Andreas Koukounas, Georgios Mastrapas, Michael Günther, Bo Wang, Scott Martens, Isabelle Mohr, Saba Sturua, Mohammad Kalim Akram, Joan Fontanals Martínez, Saahil Ognawala, Susana Guzman, Maximilian Werk, Nan Wang, Han Xiao · 2024

Related benchmarks

Task	Dataset	Result
Visual document retrieval	ViDoRe Avg. across 4 datasets v2	Full NDCG—	45
Multimodal-to-text retrieval	MM-BRIGHT	Acad Score22.3	24
Document Retrieval	DocHaystack-1000	Recall@13.67	15
Document Retrieval	InfoHaystack 100	Recall@143.23	15
Document Retrieval	InfoHaystack-1000	Recall@123.87	15
Document Retrieval	DocHaystack 200	Recall@19.17	15
Visual document retrieval	ViDoRe 8 datasets v3	NDCG@520.7	14
Visual document retrieval	ViDoRe 10 datasets v1	NDCG@553.7	14
Document Retrieval	DocHaystack 100	Recall@116.51	7
Document Retrieval	InfoHaystack 200	Recall@136.77	7