際際滷

際際滷Share a Scribd company logo

Visual Search engine with MXNet Gluon and HNSW

Download as PPTX, PDF
T
Thomas Delteil

In this talk I introduce a project I worked on, creating a Visual Search engine for 1M Amazon product using MXNet Gluon and the K-Nearest Neighbor search library HNSW. For implementation details, check this repository: https://github.com/ThomasDelteil/VisualSearch_MXNet Video available here: https://www.youtube.com/watch?v=9a8MAtfFVwI Demo website available here: https://thomasdelteil.github.io/VisualSearch_MXNet/

1 of 12
Downloaded 28 times
Thomas Delteil @tdelteil Visual Search powered by MXNet Gluon Applied Scientist | AWS Deep Engine
2AMAZON CONFIDENTIAL
3
Angel.co: 74 visual search startups Syte.ai: raised $8M Slyce.it By 2020, more than half of all mobile searches will be voice or visual searches. IPG Media Lab 4 Growing Market
How does it work? 5
6 Index images to lower-dimensional representation Images: 224x224x3 dimensions 512 dimensions finger-printpre-trained CNN (resnet18) index
7 New Image Query 224x224x3 dimensions 512 dimensions finger-printpre-trained CNN (resnet18) k-Nearest Neighbor (k=3)
Demo 8 https://thomasdelteil.github.io/VisualSearch_MXNet/
Notebook 9 https://github.com/ThomasDelteil/VisualSearch_MXNet
Demo Resource Python Notebook Apache MXNet KNN Search: Hierarchical Navigable Small Worlds (hnsw) Dataset: 8M Amazon products (2013) Image-based recommendations on styles and substitutes J. McAuley, C. Targett, J. Shi, A. van den Hengel SIGIR, 2015 10
- Develop model using a Jupyter notebook - Train model on GPU instance - Package model behind web API in a Docker container, e.g using MXNet Model Server - Upload container to container registry - Deploy container to an elastic container service / FARGATE task - Enjoy quick and linear scaling - Put the API behind a load balancer with SSL termination - Enjoy Workflow and Operationalization Elastic Container Service GPU instance Container Registry Auto-scaling Load BalancerContainer HTTPS request HTTPS response { prediction : { { item1: } }
Thanks! Feedback, follow-up and questions: tdelteil@amazon.com github.com/thomasdelteil 12

More Related Content

Visual Search engine with MXNet Gluon and HNSW

  • 1. Thomas Delteil @tdelteil Visual Search powered by MXNet Gluon Applied Scientist | AWS Deep Engine
  • 3. 3
  • 4. Angel.co: 74 visual search startups Syte.ai: raised $8M Slyce.it By 2020, more than half of all mobile searches will be voice or visual searches. IPG Media Lab 4 Growing Market
  • 5. How does it work? 5
  • 6. 6 Index images to lower-dimensional representation Images: 224x224x3 dimensions 512 dimensions finger-printpre-trained CNN (resnet18) index
  • 7. 7 New Image Query 224x224x3 dimensions 512 dimensions finger-printpre-trained CNN (resnet18) k-Nearest Neighbor (k=3)
  • 10. Demo Resource Python Notebook Apache MXNet KNN Search: Hierarchical Navigable Small Worlds (hnsw) Dataset: 8M Amazon products (2013) Image-based recommendations on styles and substitutes J. McAuley, C. Targett, J. Shi, A. van den Hengel SIGIR, 2015 10
  • 11. - Develop model using a Jupyter notebook - Train model on GPU instance - Package model behind web API in a Docker container, e.g using MXNet Model Server - Upload container to container registry - Deploy container to an elastic container service / FARGATE task - Enjoy quick and linear scaling - Put the API behind a load balancer with SSL termination - Enjoy Workflow and Operationalization Elastic Container Service GPU instance Container Registry Auto-scaling Load BalancerContainer HTTPS request HTTPS response { prediction : { { item1: } }
  • 12. Thanks! Feedback, follow-up and questions: tdelteil@amazon.com github.com/thomasdelteil 12