Understand multi-vector retrieval and how representing images as collections of patch-level embeddings enables more accurate search than single-vector methods.
Intermediate
Multi-vector Image Retrieval
Instructor: Kacper Łukawski
Earn an accomplishment with PRO · Updated Dec 10, 2025
- Intermediate
- 7 Video Lessons
- 5 Code Examples
- 1 Graded Assignment PRO
- Earn an accomplishment with PRO
- Updated Dec 10, 2025
- Instructor: Kacper Łukawski
Qdrant- Learn more aboutMembership PRO Plan
What you'll learn
Explore the ColPali model for image retrieval and apply optimization techniques like quantization and pooling to reduce memory usage while maintaining search quality.
Build a production multi-modal RAG system combining ColPali retrieval with the MUVERA model’s efficient search to handle complex documents mixing text and images.
Course recap
PROThis course teaches multi-vector image retrieval techniques, where documents and images are represented by many smaller vectors (for patches or tokens) rather than a single large vector. Built with Qdrant, it covers ColBERT for text retrieval, ColPali for image retrieval, optimization techniques for memory and search efficiency, and MUVERA which combines multi-vector and single-vector approaches.
Concept map
Key concepts
- EmbeddingsMulti-vector representations where each document/image is represented by many smaller vectors for fine-grained matching
- Semantic searchToken-level matching between query tokens and image/document patches, enabling more precise retrieval than single-vector approaches
- Retrieval-Augmented GenerationMultimodal RAG systems that retrieve images and documents based on multi-vector similarity
Lesson highlights
- 1.Introduction — Multi-vector vs single-vector retrieval and why it matters for image search
- 2.ColBERT for Text — Multi-vector text retrieval fundamentals with late interaction scoring
- 3.ColPali for Images — Adapting multi-vector approach to image retrieval using vision-language model patch embeddings
- 4.Optimization Techniques — Addressing memory and search efficiency challenges of storing many vectors per document
- 5.MUVERA — Combining multi-vector and single-vector retrieval for the best of both approaches
- 6.Building a Multimodal RAG System — End-to-end implementation with Qdrant vector database
- 7.Conclusion — Future directions for multi-vector retrieval in production systems
About this course
Join our new short course, Multi-Vector Image Retrieval! Learn from Kacper Łukawski, Senior Developer Advocate at Qdrant.
Most retrieval systems represent each image with a single vector, but multi-vector techniques represent images as collections of smaller vectors—one for each patch. This detailed representation enables fine-grained matching between text query tokens and image patches, delivering higher-quality search on complex documents that combine text, images, and diagrams.
In this course, you’ll learn multi-vector retrieval concepts, implement ColPali for image search, and apply optimization techniques to make these systems production-ready. You’ll work with real course materials to build a complete multi-modal RAG system.
In detail, you’ll:
- Learn multi-vector text retrieval with ColBERT – Understand how multi-vector retrieval works by implementing ColBERT for text, then explore the computational and memory challenges of late interaction search.
- Implement ColPali for multi-vector image retrieval – Apply multi-vector concepts to images using ColPali, which adapts vision language models to create patch-level embeddings for fine-grained visual search.
- Optimize ColPali’s memory footprint – Apply scalar and binary quantization, plus row, column, and hierarchical pooling techniques to dramatically reduce memory usage while preserving search quality.
- Enable fast search with MUVERA embeddings – Convert multi-vector representations into high-dimensional single vectors using MUVERA, unlocking HNSW search for significantly faster retrieval.
- Build a multi-modal RAG system with ColPali – Combine ColPali retrieval, memory optimizations, and MUVERA search into a complete RAG pipeline that retrieves and reasons over visual documents.
Start building retrieval systems that understand images at the patch level and deliver accurate multi-modal search.
Who should join?
AI builders working with multi-modal data who want to implement advanced image retrieval. Familiarity with Python and vector embeddings is recommended.
Course Outline
7 Lessons・5 Code Examples- IntroductionVideo・3m
- Multi-vector Text Retrieval: ColBERTVideo with Code Example・17m
- Multi-vector Image Retrieval: ColPaliVideo with Code Example・15m
- Optimizing retrieval with multi vector representationsVideo with Code Example・15m
- MUVERA EmbeddingsVideo with Code Example・18m
- Building multi-modal RAG with ColPaliVideo with Code Example・11m
- ConclusionVideo・1m
- Quiz
Graded・Quiz
・10m
Elevate your learning experience with Pro
Upgrade to Pro and gain unlimited accomplishments on your resume
Instructor
Additional learning features, such as quizzes and projects, are included with DeepLearning.AI Pro. Explore it today
Want to learn more about Generative AI?
Keep learning with updates on curated AI news, courses, and events, as well as Andrew’s thoughts from DeepLearning.AI!