Searching for Products by Photo
summary
A practical case study of integrating machine learning into production to implement visual product search for online stores, using embeddings and vector search as the core retrieval technology. Covers data preprocessing steps such as background removal and main-object extraction, and discusses the end-to-end engineering experience of deploying this ML feature.
AI-written summary · about
chapters (26)
- 0:00Introduction
- 0:49What is photo search
- 2:32Image recognition history
- 6:30Neural networks and image vectors
- 8:57Comparing image vectors
- 10:01Color in search
- 11:45Background and palette issues
- 13:52Lab and RGB color spaces
- 14:13Vector calculation infrastructure
- 15:20Vector search databases
- 16:24Faiss indexes: Flat and IVF
- 19:58HNSW indexes
- 22:46Book on indexes and implementations
- 23:07Company system architecture
- 26:45Server selection for vectors
- 28:56Cloud vs own servers
- 31:02Implementation tips and conclusions
- 31:43Q&A
- 32:04Q1: Model and vector size
- 32:45Q2: Index updates
- 33:27Q3: Search optimization
- 35:15Q4: Index actualization
- 36:17Q5: Distance algorithm choice
- 38:01Q6: Text and photo search
- 39:05Q7: Most interesting task
- 40:31Conclusion
related talks
22mHarnessing image and text data for product matching
27mMachine Learning for Advanced AI Search
33mMachine Learning in Visual Search
22mMachine Learning: The Engine Behind Modern E-commerce Search
31mAI Search for Marketplaces: Experiences & Tech
30mEvolving Search in Online Marketplaces: Practical Applications of AI
01h 26mBuilding an Image Search Engine with Generative Computer Vision Models
31mAdvanced search for your e-commerce website