“Image-service platform”

Company-wide solution of image similarity search implementing deep-learning analysis techniques, including:

  • Batch analysis and feature extraction of a billion-scaled image database
  • Building search index and search for similar images
    • Fast approximate search in billion-scaled index in 20 ms
  • Object detection
  • Face recognition
  • Attribute extraction
  • Pattern extraction for fashion items
  • Detection of
    • Frame presence
    • Background clearance
    • Text presence

Achievements

  • Architect technology solutions and software of platform.
  • Automation of indexing images, training models, updating indexes using Hadoop and Spark on top of Mesos. 
  • Architect, develop and deploy:
    • highly-scalable micro-service based architecture of search component.
    • fully-automated index building  including incremental injections.
    • similarity-search project for billion-scale index
    • visual features extractor for online and batch processes
    • API
    • in-house demo website
  • Built and led engineering team
  • 200x increased performance of image-similarity searcher by implementing fast LOPQ search
  • Viber bot to search images similar to captured from smartphone camera
  • CI/CD always
Posted in Project

Leave a Reply