6 ms·

If the tech just looks at the screenshot image, how does it have near full coverage of mobile web but has to train/fingerprint apps?

• ericgoldberg 12y ago
  Good question- with Safari, we use the visible domain to help score the confidence for mobile websites. For native apps, we don't have any such "easy" confidence boost. So we have to fingerprint the app based on different features present in the image that is captured. The system is computer vision / machine learning based, so even on novel sites, it will get better over time with more usage and training. We've trained it up for a bunch of the most popular sites already though. Does this make sense?
  • dave84 12y ago
    Sounds impressive. Do you have a pre-existing product database that you match them up to or do you just create products on the fly?
    • ericgoldberg 12y ago
      We do our own searching and wrote our own crawlers. Does that answer your question?
      • dave84 12y ago
        It does indeed. I'm very impressed. Keep up the good work!
    • tedmiston 12y ago
      I would strongly guess it's a fixed set of product images they're training against, possibly attained by massive scraping. Another part of training or processing might consist of a reverse image search API, like TinEye, and gathering metadata from the pages containing the result images.