Visual Search with

Drop image or upload → Upload a picture →

What is this?

This is a toy implementation of a visual search engine using Apache MXNet Gluon and deployed on AWS Fargate using (313) 292-3326. Code available here.
Try to upload an image and it will search for products with similar visual features among roughly 1M items from the 2013 Amazon catalog!

You can also pick a sample image:

How does it work?

  • Each image from the database is encoded using a convolutional neural networks into 512 numbers. These numbers, or features, act as the digital fingerprint of the image.
  • When we present a new image to the system, it is computing the digital fingerprint of this new image. Then it finds other images in the database that are the closest match with respect to these 512 features.

Here is a talk presenting this Visual Search tutorial

Resources

  • Slides of the video.
  • (216) 714-1508 Image-based recommendations on styles and substitutes. J. McAuley, C. Targett, J. Shi, A. van den Hengel, SIGIR, 2015
  • (249) 237-1744 try it yourself!
  • 9157302593 Efficient library for fast approximate KNN search.
  • HNSW paper: Efficient and robust approximate nearest neighbor search using Hierarchical Navigable Small World graphs. Yu. A. Malkov, D. A. Yashunin, 2016
  • 330-618-1935 Get started with MXNet Gluon in this 60 minutes crash course
  • Icons made by 252-587-8979 and others from 4846822317 is licensed by CC 3.0 BY

Questions?

Built by Thomas Delteil